Charge Sensing and Spin Dynamicsin GaAs Quantum Dots
A thesis presentedby
Alexander Comstock Johnson
toThe Department of Physics
in partial fulfillment of the requirementsfor the degree of
Doctor of Philosophyin the subject of
Physics
Harvard UniversityCambridge, Massachusetts
September 2005
c© 2005 by Alexander C. JohnsonAll rights reserved.
Thesis Advisor: Author:
Charles M. Marcus Alex C. Johnson
Charge Sensing and Spin Dynamics
in GaAs Quantum Dots
Abstract
This thesis describes a series of experiments which explore and the technique of charge
sensing in GaAs/AlGaAs quantum dot devices, then use charge sensing as a tool to
investigate spin and charge dynamics. The first experiment generates Fano resonances
and shows that charge sensing modifies the resonances in nontrivial ways. Charge
sensing is also responsible on its own for features similar to Fano resonances. A
phenomenological model is proposed and is is good quantitative agreement with the
measured lineshapes.
The remaining experiments use sensing to measure charge in a double quantum dot.
Equilibrium measurements (no source-drain bias or voltage pulses) are presented,
which allow sensitive determination of interdot tunnel coupling and electron tem-
perature. Adding a source-drain bias dramatically illuminates the spin properties of
few-electron double dots via the phenomenon of singlet-triplet spin blockade, visible
in both transport and sensing measurements.
By adding gate-voltage pulses to these systems, we can measure energy spectra and
dynamics in isolated or nearly-isolated double dots. In the isolated case we demon-
strate excited state spectroscopy and single-shot charge position measurements. In
iii
few-electron double dots we present spin relaxation and dephasing measurements
showing that the effective magnetic field from hyperfine interaction with the host Ga
and As nuclei causes random spin precession on a timescale of ∼10 ns (measured both
by direct time-domain measurements and by a linewidth in field), and that this is the
only significant source of spin dephasing up to ∼1 ms.
iv
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
1 Introduction 1
1.1 Organization of this thesis . . . . . . . . . . . . . . . . . . . . . . . . 3
2 GaAs nanostructures 6
2.1 GaAs/AlGaAs heterostructures . . . . . . . . . . . . . . . . . . . . . 7
2.2 Contacting the 2DEG . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Depletion gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Quantum transport 14
3.1 Quantum point contacts . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Quantum dots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Double dots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Charge sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
v
4 Coulomb-modified Fano resonance in a one-lead quantum dot 36
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Device behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Model for Coulomb-modified Fano resonance . . . . . . . . . . . . . . 43
4.4 Analysis and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5 Differential charge sensing and charge delocalization in a tunabledouble quantum dot 50
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Charge sensing honeycombs . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Temperature and tunnel coupling . . . . . . . . . . . . . . . . . . . . 56
6 Charge sensing of excited states in an isolated double quantum dot 62
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.2 Tunnel-coupled and isolated double dots . . . . . . . . . . . . . . . . 66
6.3 Single-pulse technique . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.4 Latched detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7 Singlet-triplet spin blockade and charge sensing in a few-electrondouble quantum dot 74
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.2 Spin blockade at (1,1)-(0,2) . . . . . . . . . . . . . . . . . . . . . . . 78
7.3 Spin blockade at other charge transitions . . . . . . . . . . . . . . . . 87
8 Triplet-singlet spin relaxation via nuclei in a double quantum dot 92
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
vi
8.2 Pulsed-gating technique . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.3 Spin relaxation measurements . . . . . . . . . . . . . . . . . . . . . . 97
8.4 Analysis and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9 Singlet separation and dephasing in a few-electron double quantumdot 106
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
9.2 The spin funnel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
9.3 Measurement and theory of T ∗2 . . . . . . . . . . . . . . . . . . . . . . 113
A Hyperfine-driven spin relaxation 116
A.1 Definition of Bnuc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
A.2 Hyperfine-driven decay . . . . . . . . . . . . . . . . . . . . . . . . . . 118
A.3 Thermal component . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
B Electronics and wiring 122
B.1 DC and RF wiring in a dilution fridge . . . . . . . . . . . . . . . . . 122
B.2 AC+DC box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
B.3 Grounding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
B.4 Divider/adder for NI6703 DAC . . . . . . . . . . . . . . . . . . . . . 128
C Cold finger design 130
D Igor routines 134
E Measurement techniques 144
E.1 Cooling and diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . 144
vii
E.1.1 Room-temperature sample tests . . . . . . . . . . . . . . . . . 146
E.1.2 Cooling with positive bias . . . . . . . . . . . . . . . . . . . . 149
E.1.3 Low-temperature sample tests . . . . . . . . . . . . . . . . . . 150
E.2 Wall control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
E.2.1 Linear charge sensor control . . . . . . . . . . . . . . . . . . . 151
E.2.2 Automatic charge sensing . . . . . . . . . . . . . . . . . . . . 152
E.2.3 Honeycomb centering . . . . . . . . . . . . . . . . . . . . . . . 153
E.2.4 Matrix wall control . . . . . . . . . . . . . . . . . . . . . . . . 153
viii
List of Figures
2.1 Wafer structure of a 2DEG . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Depletion gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Multi-scale view of a chip . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1 Sample QPC conductance traces . . . . . . . . . . . . . . . . . . . . . 19
3.2 Schematic description of a quantum dot . . . . . . . . . . . . . . . . 24
3.3 Schematic description of a double quantum dot . . . . . . . . . . . . 29
4.1 Channel conductance in the Fano regime . . . . . . . . . . . . . . . . 38
4.2 Three configurations with a tunnel coupled dot . . . . . . . . . . . . 40
4.3 Charge sensing by a channel resonance . . . . . . . . . . . . . . . . . 42
4.4 Experimental Fano resonances and fits . . . . . . . . . . . . . . . . . 47
5.1 SEM micrograph of the device . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Conductance and sensing honeycombs . . . . . . . . . . . . . . . . . . 54
5.3 Sensor signal along a detuning diagonal vs. temperature . . . . . . . . 57
5.4 Tunnel coupling dependence of sensor signal . . . . . . . . . . . . . . 60
6.1 Device and measurement regimes . . . . . . . . . . . . . . . . . . . . 65
6.2 Single-pulse technique . . . . . . . . . . . . . . . . . . . . . . . . . . 68
ix
6.3 Detailed analysis of single-pulse technique . . . . . . . . . . . . . . . 70
6.4 Two-pulse technique . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.1 Spin blockade device and schematics . . . . . . . . . . . . . . . . . . 76
7.2 Spin blockade in transport . . . . . . . . . . . . . . . . . . . . . . . . 79
7.3 Spin blockade in sensing . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.4 Measurements of the singlet-triplet splitting . . . . . . . . . . . . . . 86
7.5 Spin blockade at other charge transitions . . . . . . . . . . . . . . . . 88
8.1 Spin-selective tunnelling in a double quantum dot . . . . . . . . . . . 94
8.2 Gate pulse calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.3 grs as a function of VR and VL for increasing B . . . . . . . . . . . . . 98
8.4 Dependence of the occupancy of the (1,1) state on measurement timeand external field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
8.5 Detailed measurements of blockaded (1,1) occupation . . . . . . . . . 101
8.6 Decay of (1,1) occupancy as a function of detuning at various magneticfields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
9.1 Measurement scheme for T ∗2 . . . . . . . . . . . . . . . . . . . . . . . 109
9.2 Spin dephasing in a double dot . . . . . . . . . . . . . . . . . . . . . 111
9.3 Composite plot of P(1,1) vs. εs and B . . . . . . . . . . . . . . . . . . 112
9.4 P(1,1) measured as a function of τs . . . . . . . . . . . . . . . . . . . . 114
B.1 DC and RF wiring in the Nahum fridge . . . . . . . . . . . . . . . . . 124
B.2 AC+DC box schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 125
B.3 Full grounding configuration for an experiment . . . . . . . . . . . . . 127
x
B.4 Divider/adder circuit for NI6703 . . . . . . . . . . . . . . . . . . . . . 129
C.1 Schematic and photograph of cold finger . . . . . . . . . . . . . . . . 132
E.1 Chip carrier for testing fridge wiring . . . . . . . . . . . . . . . . . . 145
E.2 Gate/ohmic test circuit . . . . . . . . . . . . . . . . . . . . . . . . . . 146
xi
Acknowledgements
Experimental physics is an inherently collaborative effort on just about all levels,
all the way from talks with experts from around the world, where one parenthetical
remark can spark a new line of experiments, to times when I might have forgotten
to eat or sleep without a little help from my friends. There are far too many people
who contributed to my Ph.D. for me to have any hope of thanking them all, but here
we go.
My first thanks go to Charlie, without whose wealth of experience, unending stream
of ideas, and remarkable intuition, none of this would have been possible. I am
particularly grateful to Charlie for his sincere passion for the truth and his devotion
to doing things right, in lab, in publications, and in the world beyond physics.
The other grad students in the group are the ones who make the day-to-day expe-
rience what it is. My time here began with many late nights with Ron, Josh, and
Andrew, and eventful days with Sara, Jeff, Dominik, and Heather. Despite the failing
experiment, Josh teaching me and Ron to run a fridge was a great introduction to
low-T physics. Ron was always happy to talk about physics, especially when he had a
paper to write, and Josh was always happy to plan the next experiment (or hike, bike
ride, or ski trip). With Andrew around there was never a dull moment, and always a
new game to play (crossword, foam-based, or otherwise). Sara’s no-nonsense attitude
injected a healthy dose of realism and perspective into Marcus lab life, and the lab has
never really recovered from the departure of its social coordinator. Heather taught
xii
us all what a lab notebook could look like, and Dominik showed that if you take
twelve papers worth of data at once you’ll have more time for parties and dancing.
Jeff, besides brightening the lab with his incredible creativity (and his award-winning
plants), could make a photo album of everything that happened during my tenure
in lab, which he is the only one to share in its entirety. Jeff’s easy-going but orga-
nized manner meant that when he was around you knew everything would work out,
because there was always a backup plan.
The next year Leo and Mike joined the lab, increasing the lab’s diversity of expertise
(Leo the electronics and RF jock, and Mike the nanotube jock) and personalities
(Leo the crazy latin drummer turned marathon runner and Mike the metro DJ poli-
tics junkie). I owe many thanks to deli meat for resuscitating me after brutal penny
wars with Mike. Working with Jason over the last year and a half has been a real
pleasure. His manner of powering through any problem is inspiring, be it in fabrica-
tion, measurement, or an effeminate piece of hardware. I’d also like to thank Jake
and Lily, who managed to straddle the worlds of theory and experiment admirably,
Edward who is just as quick a learner in lab as on the slopes, and Yiming who almost
seemed cheerful about taking the evaporator job from me. I could go on, thanking
all of the other students, postdocs, visitors, and the excellent staff at Harvard (the
group would grind to a halt without the able help of James I, II, and III, Tomas and
Matt, Jim MacArthur, Louis and his machinists, Ralph and Joan, and many others)
but I have the feeling this acknowledgements section would never end.
So finally I’d like to thank Stela for bringing more joy than she knows to my life, and
my parents for making this all possible, encouraging and supporting me without fail
through more than twenty years of formal education.
xiii
Chapter 1
Introduction
The theory of quantum mechanics is one of the great success stories of 20th century
science, and it goes hand in hand with the explosion of electronics technology. As
dramatic as this technological advance has been, however, so far most electronic
devices have been engineered semiclassically: quantum mechanics tells us what kind
of potential electrons feel in different materials but beyond that the electrons are
treated as classical particles. There is still a long way to go before the size and
energy scales of individual atoms limit the miniaturization and speed of electronics.
While the electronics industry tends to take a top-down approach, starting with
large, well-understood systems and watching what happens upon shrinking them,
basic physics, especially theoretical physics, tends to prefer a bottom-up approach,
building complex systems from their well-understood component particles. In one
trivial sense the quantum mechanics of everyday particles is complete: the equations
of motion for electrons and atomic nuclei interacting with each other are known, so
in principle we can predict what will happen to any given collection of such particles
in any given initial state. But while simulations based on the exact equations are
1
often useful in classical physics—for designing cars and bridges, for sending rockets
to other planets—the complexity of the exact equations grows so much more quickly
in quantum than in classical mechanics that simplifying assumptions must be made
for even a small number of particles. Grappling with this complexity is the central
pursuit of condensed matter physics.
In the last several decades, thanks mainly to advances in nanoscale fabrication tech-
niques, a new research field dubbed mesoscopic physics has emerged, at the boundary
of these top-down and bottom-up understandings of many-particle systems. Meso-
scopic (meaning medium-sized) refers to the length scale of particle confinement being
smaller than its coherence length, such that the localized effects of quantum mechan-
ics become important, but not as small as the atomic scale at which it would interact
with only a few other particles.1
This thesis concerns electrons confined to a 2-dimensional electron gas at the bound-
ary of two semiconductors, GaAs and AlxGa1−xAs. This system has been remarkably
prolific because of the ease of engineering a wide variety of quantum states. The
experiments presented here are linked by a technique known as charge sensing, which
has been known for over a decade but only in the past few years has its power and
versatility become apparent. This thesis helps develop charge sensing from a tech-
nique under investigation in its own right to a robust method of measuring isolated
systems, which yields otherwise inaccessible information about the energy spectra,
the spin states, and the dynamics of electrons in a semiconductor lattice.
1Some definitions limit mesoscopics to systems of particles confined on scales largerthan their wavelengths [1], such that they interact with many of each other. As wewill see, however, interesting many-particle interaction effects are seen with even asingle electron if its wavefunction is spread over many lattice sites.
2
1.1 Organization of this thesis
Chapters 2 and 3 review the background necessary to understand the experiments in
this thesis. Chapter 2 introduces GaAs nanostructures and the techniques for creating
and measuring them, and Ch. 3 discusses the main concepts involved in understanding
the particular nanostructures used in the experiments to follow, namely quantum
point contacts and single and double quantum dots.
Chapter 4 discusses an experiment designed to measure Fano resonances. The analy-
sis of these data constituted a watershed moment in my thinking about quantum dot
measurements; though other members of the group were already working on charge
sensing, the Fano resonance experiment showed the ubiquity and strength of charge
sensing, to the extent that while Fano resonance could be found in certain small
portions of parameter space, charge sensing was, to our dismay at the time, reliably
found everywhere. Connect that reliability with the other advantages we were begin-
ning to realize charge sensing possessed, and sensing became the new tool of choice.
Chapter 5 describes the use of charge sensing in a double quantum dot to determine
not just the presence but the positions of electrons. This allowed us to study isolated
double quantum dot devices, which never exchange charge with any reservoirs but
merely shuttle electrons from one side to the other. Chapter 6 describes pulsed-gate
measurements of such an isolated device, yielding the excited state spectrum and
allowing a single-shot measurement of charge location.
Next we move to a smaller double dot design, which can be tuned to zero, one, or
two electrons in either dot with arbitrary interdot and dot-lead tunnel couplings. In
this few-electron regime, orbital energy level separations become very large while the
3
available spin states become simple and well-understood. Chapter 7 describes DC
transport and charge sensing measurements of this device in both the one-electron
and two electron cases. A strong bias asymmetry is observed in the two-electron case,
which fits neatly into a model in which the ground state for two electrons in one dot
is a spin singlet whereas two separated electrons may take either a singlet or a triplet
spin state. This effect, known as singlet-triplet spin blockade, formed the basis of
our subsequent pulsed-gate measurements investigating two-electron spin dynamics
in real time. Chapter 8 describes our first successful pulsed-gate scheme in this device.
In this scheme, a separated two-electron state is prepared with a uniform mixture of
spin states, then tilted so that the ground state has both electrons in the same dot.
As this experiment was performed with weak interdot tunneling, it tells us about
the long-time behavior of spins in this system, and the magnetic field dependence of
the results pointed to hyperfine interaction with nuclei as the main source of spin
evolution. Chapter 9 describes the second successful pulsed-gate scheme, in which
a separated singlet state is prepared by splitting the overlapping singlet, then after
some time we attempt to push the electrons back into the same dot. The time
dependence of these results give us the first direct measure of the inhomogeneous
spin dephasing time T∗2 in a single-electron quantum dot, while the magnetic field
dependence provides insight into the energy levels of this system with unprecedented
precision.
The appendices provide some experimental and theoretical details which I hope will be
useful to others in the field. Appendix A provides more detailed theory of quasistatic
nuclear fields, backing up Ch. 8, much of which is relevant also to Ch. 9. Appendix B
describes some of the circuitry and wiring I built for use inside and outside the
4
cryostat, and Ap. C describes our cold finger design. Appendix D describes some
Igor programs I developed for data acquisition and management. Finally, Ap. E has
some miscellaneous techniques for measuring devices efficiently.
5
Chapter 2
GaAs nanostructures
Quantum state engineering involves the creation of well-defined potential profiles on
the scale of the partcles’ wavelengths. The known ways of doing this are myriad.
Photons and atoms can interact in free space [2, 3] or solid state [4, 5] cavities.
Electrons (or composite particles such as excitons or cooper pairs) can be bound to
nanostructured materials such as carbon nanotubes [6, 7], semiconductor nanowires
[8], metal grains [9, 10], self-assembled quantum dots [11, 12], or lithographically
patterned superconductors [13, 14] or semiconductors.
Of all these techniques, perhaps none has shown the versatility and tunability of
the lithographically patterned GaAs/AlGaAs semiconductor heterostructure. Two-
dimensional systems can be made with unrivaled homogeneity and arbitrary density
[15, 16, 17]. From this starting point, one-dimensional [18, 19, 20, 21] and fully quan-
tized zero-dimensional systems [22, 23, 24, 25] can be made in almost any imaginable
combination, and the individual elements of the Hamiltonian—energy levels, tun-
nel couplings, interference phases, and more—can be tuned by external parameters
within a single device. This chapter describes some of the techniques for creating and
6
10 nm - GaAs cap
60 nm - Al0.3Ga0.7As
4 x 1012 cm-2 - Si δ-doping
40 nm - Al0.3Ga0.7As
800 nm - GaAs
30-period superlattice3 nm Al0.3Ga0.7As, 3 nm GaAs
500 µm - GaAs substrateenergy
2DEGferm
i level
Figure 2.1: Wafer structure of a 2DEG. At right the energy of the conduction bandedge is shown schematically vs. depth. A trianglular potential well is formed at theburied GaAs/AlGaAs interface with one subband (thick solid line) below the Fermilevel. Other subbands (thick dashed line) are inaccessible. The specific parametersshown here describe the nominally identical wafers 010219B and 031104B, grown byMicah Hanson at UCSB, which were used to make all of the devices used in theexperiments in this thesis.
measuring this endless variety of devices.
2.1 GaAs/AlGaAs heterostructures
The first step in creating a GaAs nanostructure is to make a two-dimensional electron
gas (2DEG) [26] at the interface between GaAs and AlxGa1−xAs (typically x ∼ 0.3)
as shown in Fig. 2.1. Positively charged donors (usually group IV silicon atoms
substituting for group III gallium or aluminum) are placed tens of nanometers away
from the interface in the AlGaAs region. Because AlGaAs has a larger band gap
than GaAs, the global potential minimum is not at the donors but at the interface
7
with the GaAs. This separation of dopants and charge carriers is called modulation
doping, and is responsible for the extraordinarily smooth potential in the 2DEG
plane in two ways. First, because GaAs and AlGaAs have nearly identical lattice
constants and can be grown with atomic precision using molecular beam epitaxy
(MBE), there are few defects at the interface. Second, potential fluctuations due to
the random locations of the donors are smoothed by their distance from the interface
(40 nm for these wafers), which is larger than the average distance between donors
(5 nm). Modulation-doped heterostructures can achieve mean free paths as high as
100 µm even for densities as low as one electron per (100 nm)2. The other layers in a
heterostructure wafer play subtler roles in maintaining 2DEG quality: far below the
2DEG there is a GaAs/AlGaAs superlattice which smoothes any impurity potential
in the underlying bulk GaAs wafer, and above the AlGaAs layer is a thin cap of GaAs,
to prevent oxidation of the surface, which could also occur randomly and affect the
potential at the 2DEG.
2.2 Contacting the 2DEG
All of the measurement techniques I will describe involve measuring the current/vol-
tage characteristics of various regions of 2DEG, and to do this we need to make
electrical contact between the 2DEG and a metal, that we can attach a wire to. These
are referred to as ohmic contacts because good ones behave just like small resistors.
Two barriers must be overcome to make contact to a 2DEG: the structural barrier
(the 100 nm insulating AlGaAs layer above the 2DEG) and the Schottky barrier
inherent in a metal-semiconductor interface. The most common contact method, and
8
the only one used in the devices in this thesis, is to deposit gold-germanium on the
surface and anneal it into the wafer. The standard picture invoked to describe the
annealing process is that spikes of gold extend through the AlGaAs to penetrate the
physical barrier, bringing germanium to diffuse into the GaAs, where it is an n-type
dopant and thus lowers the Schottky barrier. Within this model it is easy to explain
the need to precisely tune the annealing time and temperature: too little annealing
and there will be too few metal spikes reaching the 2DEG layer; too much annealing
and the spikes will all merge, such that electrons may only enter the 2DEG at the
very edge of the contact. For this reason, every time a new wafer is used, a new
annealing recipe must be determined by trial and error for best contact.
The following is a typical ohmic contact recipe: Deposit 5 nm Pt, 100 nm AuGe, 25
nm Pt, and 100 nm Au on the clean surface of the wafer, then heat to 480 C for 60
seconds. The first platinum layer is very thin, and is present solely to help the gold-
germanium stick to the surface. In the past the group has used nickel for this purpose,
but we have avoided this more recently because nickel is ferromagnetic and causes
heating when the magnetic field is swept through zero.1 The gold-germanium layer
that follows is intended to be a eutectic (lowest melting point) mixture of 88% gold,
12% germanium. The eutectic temperature is 356 C (compared with ∼1000 C for
either pure gold or pure germanium). This is well below the annealing temperature, so
this layer liquefies during the anneal, which presumably greatly enhances its diffusion
into the wafer. Above the AuGe is a platinum barrier layer and then a layer of pure
gold to aid wirebonding. In practice, however, another layer of gold is often deposited
1It isn’t clear which bit of nickel was responsible for this heating. We eliminatednickel here and in the sticking layer for gold-plated connectors in several locationsnear the device, and this field-sweep heating has disappeared.
9
after annealing because this first bonding layer can harden during the anneal.
A good contact has two closely related properties: it follows Ohm’s law V = IR over
the useful current range |I| < 100nA,2 and the resistance R of the contact is small
compared to the device to be measured. For most measurements of GaAs nanos-
tructures, R < 1kΩ is considered good, although we have on occasion seen contact
resistances of 50 Ω or less for normal ∼ 100-150 µm square pads. Contacts with resis-
tance R > 10kΩ can still be used if necessary, but they are likely to show pronounced
non-Ohmic and frequency-dependent response, which complicates the interpretation
of data, even if used in a four-wire configuration which normally removes the contact
resistance from the measurement.
2.3 Depletion gates
Nanostructures can be created in an existing 2DEG by a variety of methods, including
etching [27], local oxidation [28, 29], local charging [30], and electrostatic gating
[18, 31, 22, 23]. The common feature of all of these approaches is that they raise
the Fermi level in a particular region of 2DEG above the first z-subband minimum,
excluding all electrons from that region. The most versatile approach is electrostatic
gating, where electron beam lithography is used to define metal gates with feature
sizes of 100 nm or less.3 An arbitrary negative voltage can be applied to each gate,
2100 nA is the current generated by 1 mV across an open single-mode quantumchannel, R = h/2e2, as described in Sec. 3.1.
3An argument could be made that local charging by a scanned probe is moreversatile, because after cooling you can draw any arbitrarily complex device withoutextra gates, however the hysteretic nature of such charging and the inability to changemore than one parameter without moving the probe limit the use of this techniquein practice.
10
VL VR
Iin
Iout
2DEG
Figure 2.2: Metal gates are deposited on the surface of the wafer, defined by electronbeam lithography. Wires are bonded to these gates and attached to external negativevoltages, labeled VL and VR on this simple drawing. Regions where electrons remainin the underlying 2DEG are shown in light grey. Ohmic contact is made to the 2DEGregions on either side of the device to allow current to flow in and out of the 2DEGthrough the device.
raising the potential and removing the electrons from a variably sized region under
and around the gates, as illustrated in Fig. 2.2. Determining exactly what potential
is created in the 2DEG for a specific configuration of gates and gate voltages can
be extremely difficult for a number of reasons.4 Fortunately the exact details of
the potential are seldom important, and most systems can be expressed by effective
Hamiltonians in which the important parameters can be measured experimentally
and tuned by the gate voltages.
4For example, the electrons in the 2DEG screen the gate potential to some degree,but due to their long wavelengths such screening is imperfect, and stops altogetherwhere the 2DEG is depleted. Other factors complicating this analysis include the“impurity” potential due to the random locations of ionized donors or crystal impuri-ties, and selective ionization of donors due to our trick of cooling with a positive gatebias. Among others, Mike Stopa [32, 33, 34] has made a career of calculating 2DEGpotentials, and the Heller and Westervelt groups [35, 36] have extensively studiedtransport through impurity potentials.
11
1 mm
100 µm
25 µm
1 µm
5 µm
Gate pad
Mesa
Ohmic contact
Figure 2.3: Multi-scale view of Gorgonzola (aka HJL18), the chip used in Chs. 5 and6. The upper image shows the entire chip. Dashed line highlights the outline of onemesa. Bond pads on the mesa are ohmic contacts, bond pads off mesa are gates.Continuing counter-clockwise, we zoom in on the center of one device by about afactor of 5 each time. The three largest-scale images are from an optical microscope,the two smallest from a scanning electron microscope.
12
The upper image of Fig. 2.3 shows a complete chip. Because a lot of work goes
into making a good 2DEG wafer and learning how to make good ohmic contacts
and gates on it, and because there can be a significant failure rate for devices due to
fabrication errors or for unknown reasons, it is advantageous to maximize the number
of devices made out of one wafer. For this reason, we pack many devices onto a single
chip. Putting multiple devices on one chip requires an additional fabrication step
to electrically isolate the devices from each other. We do this by etching away the
2DEG between the devices, leaving behind a “mesa” of 2DEG for each device, one
of which is outlined on the upper image of Fig. 2.3. Surrounding each device is a
set of metal pads (100–150 µm squares) for wire bonding. The pads on-mesa are
ohmic contacts, and the pads off-mesa are gates. The remaining images in Fig. 2.3
magnify the center of the device, following the gates and 2DEG reservoirs from their
macroscopic bond pads down to their fine tips defining the submicron-scale device.
In the most zoomed-in image (center right), imagining a depletion halo around each
gate shows where four confined regions of electrons (quantum dots) will be located,
connected to each other by three tunnel barriers and connected to 2DEG reservoirs
by another six barriers.
13
Chapter 3
Quantum transport
This chapter lays the theoretical groundwork for the experiments described later
on. I have not tried to give an exhaustive treatment of quantum transport, rather
to introduce the physics of key device elements and give references to neighboring
applications in the literature. Most of the concepts in this chapter can also be found
in review articles or books, my favorites being Datta [37] and Imry [38] for 2D and
1D systems and scattering and the Kouwenhoven review [39] for dots in the Coulomb
blockade. Van der Wiel’s review [40] is an excellent detailed reference on Coulomb-
blockaded double dots.
3.1 Quantum point contacts
The simplest nanostructure to make in a 2DEG is a constriction, such as is shown
in Fig. 2.2, by depleting two gates with a gap between them. A short constriction is
known as a quantum point contact (QPC). A remarkable property of a QPC, and one
that was by no means obvious before it was observed, is that even if there are only a
few electron wavelengths from the entrance to the exit of the constriction it behaves
14
almost identically to the theory for an infinitely long clean wire of uniform width.1
Analyzing the infinite wire is straightforward: regardless of the shape of the transverse
potential, motion in this direction can be quantized into discrete transverse subbands,
just as was already done in the z direction due to the heterostructure. Electrons then
propagate freely along the wire with a dispersion relation E = ~2k2/2m∗ determined
solely by the effective mass m∗, which in GaAs is 0.067me, where me is the electron
mass, 9.11 × 10−31kg. As it turns out, however, we don’t even need to know the
dispersion relation to determine the conductance of this system.
Consider for simplicity the case of zero temperature and one transverse subband oc-
cupied at the Fermi energy at each end of the wire, but with higher Fermi energy on
one side (the source) than on the other side (the drain). At any energy where states
on both sides are occupied, the current passing one way through the constriction must
exactly cancel the current passing the opposite direction in order to satisfy detailed
balance. The net current is therefore given by the rate of electrons passing through
the constriction above the drain Fermi level but below the source, I = e∫ Es
EddE ∂2N
∂E∂t.
Taking the derivative with respect to voltage by using E = eV , the differential con-
ductance is
G =dI
dV= e2
∂2N
∂E∂x
∂x
∂t. (3.1)
We can evaluate the first term (the linear density of states) by applying infinite
boundary conditions to our wire and counting nodes, which gives ∂N∂x
= limx→∞Nx
=
gsx/(λ/2)
x= 4πgsk where the spin degeneracy gs = 2 for our spin-1/2 electrons at low
magnetic field. For the purposes of calculating current we consider only half of these
1Note that because the 2DEG depth (∼ 100nm) is typically several wavelengths(∼ 50nm), nearly every constriction meets this requirement, since the potential fromsurface gates is softened on roughly the scale of the depth.
15
states (those moving from source to drain), so ∂2N∂E∂x
= 2πgs∂k∂E
. The second term in
Eq. 3.1, ∂x∂t
, is the group velocity, also written as ∂ω∂k
= 1~
∂E∂k
. Putting these together,
the two derivatives exactly cancel, leaving
G = gse2
h. (3.2)
Because of the remarkable simplicity of this equation and the fact that so few as-
sumptions went into it, e2/h is known as the universal quantum of conductance.
Temperature and higher subbands can easily be added to this picture. Each time a
new subband edge crosses the Fermi level the conductance increases by gse2/h, re-
sulting in a series of conductance steps as a function of gate voltage. Temperature
has the effect of averaging over a distribution of energies, and as such it has no effect
on the levels of the conductance plateaus. The primary effect of temperature is to
broaden the transitions from sharp jumps as each subband minimum changes from
empty to filled, to smooth increases as the probability of occupying the bottom of a
new subband gradually increases from 0 to 1. In high-quality QPCs this broadening
can be used to measure electron temperature.
QPCs vary greatly in the degree to which they match this model. There is, of
course, the “0.7 structure,” studied extensively by the Marcus group and others
[41, 27, 20, 42], which affects details of the transition from zero to one mode. Here
we are concerned with more drastic effects which introduce backscattering, where
electrons which would otherwise have passed through the constriction are scattered
back to the source reservoir, altering the basic picture of smooth transitions between
conductance plateaus. This can be caused by impurities in or near the channel, by
too-sharp edges of the QPCs making the transition from 2D reservoirs to the 1D
16
channel non-adiabatic,2 by other depletion gates if the QPC is part of a more com-
plex device, or by a combination of small-angle impurity scattering and scattering off
other depletion gates. For this reason, a more complete description including a finite,
energy-dependent transmission coefficient Tn(E) for each mode n is often useful. We
can then write the conductance as
G =e2
h
∑n
Tn(Ef ) (3.3)
or, at finite temperature,
G =e2
h
∫dE
∂f(E − Ef , T )
∂E
∑n
Tn(E), (3.4)
where f(ε, T ) is the Fermi distribution for temperature T . This is known as the
two-terminal Landauer formula. Each transmission coefficient may have complicated
non-monotonic behavior as a function of energy, although in general they all tend
from zero at low energy to unity (full transmission) at high energy.
Another way to interpret a transmission coefficient less than 1 is as the onset of
tunneling. The term tunneling is usually reserved for a process which is classically
disallowed, meaning that the potential profile rises above the total energy of the
particle at some point in the barrier. A particle has negative kinetic energy while
tunneling, whereas scattering can occur even due to a drop in potential and an increase
in kinetic energy. Truly infinite 1D channels would not allow tunneling, because a
barrier of any height has zero transmission at infinite length. But real QPCs, with
lengths of only a few electron wavelengths, make very good tunnel barriers. In the end,
there’s no difference between low transmission caused by tunneling, by scattering, or
even by interference [43].
2This effect is analogous to the well-known effect from introductory quantum me-chanics, in which a particle moving in 1D can be backscattered by a sharp drop inpotential, which does not occur in classical mechanics.
17
Figure 3.1 shows some of the behaviors possible in a QPC. The QPCs in (a)–(c)
were made as independent devices, with no other nearby gates and with shapes op-
timized for clean transmission. All of these devices show multiple clear plateaus at
the expected conductance values (note that the vertical scale in (a)–(c) is in units
of 2e2/h), although in some cases there are bumps or dips in the plateaus. Contrast
this with QPCs integrated into more complex devices, with constrained geometries.
Conductance traces for four such QPCs are shown in Fig. 3.1 (d)–(g). In these QPCs,
instead of a smooth rise to a plateau at 2e2/h, we see gradual bumpy increases, often
with sharp peaks indicative of resonance, sometimes with what looks like a plateau
but at an arbitrary conductance. Behavior of these QPCs at and above 2e2/h is not
shown, but they only sometimes have a plateau at all, and generally the conductance
just continues its bumpy rise higher and higher. All of these effects indicate complex
behavior of the transmission coefficients, rather than a simple turn-on as the subband
minimum crosses the Fermi energy.
3.2 Quantum dots
Whereas a quantum point contact is the restriction of a 2D electron gas in one more
dimension, forming a 1D channel, a quantum dot is the restriction of electrons in all
three dimensions, and as such is sometimes referred to as a zero-dimensional (0D)
system. Typically, in order to study quantum dots, two QPCs are incorporated into
the perimeter of the dot, and properties of the dot are inferred from the conductance of
electrons going in one lead and coming out the other. Although many other geometries
and measurement techniques are possible (some of which are explored in the coming
18
5
4
3
2
1
0
e2( g2
)h/
-900 -700 -500 -300Vg (mV)
SMC 4c)5
4
3
2
1
0
e2( g2
)h/
-1000 -750 -500 -250Vg (mV)
SMC 2a)5
4
3
2
1
0
e2( g2
)h/-800 -600 -400 -200
Vg (mV)
SMC 3b)
d) e)2.0
1.5
1.0
0.5
g (e
2 /h)
-800 -750 -700 -650Vg (mV)
1.2
1.0
0.8
0.6
0.4
0.2
g (e
2 /h)
-850 -800 -750 -700Vg (mV)
g)1.2
1.0
0.8
0.6
0.4
0.2
g (e
2 /h)
-800 -700 -600Vg (mV)
2
1.5
1.0
0.5
g (e
/h)
-1200 -1100 -1000Vg (mV)
f)
1µm
1µm1µm
Figure 3.1: Sample QPC conductance traces. a)–c) Conductance vs. gate voltage Vg
for three QPCs made for independent study (data courtesy of S. M. Cronenwett [1]).d)–g) Analogous traces (showing only the region below 2e2/h) for four QPCs in twodevices used as charge sensors in Chs. 5 and 6 (d,e) and in Chs. 7–9 (f,g).
chapters), we begin by reviewing standard transport through a dot. Understanding
the successes and limitations of transport measurements motivates the use of charge
sensing for certain purposes, which I describe below in Sec. 3.4.
Quantum dots can be divided into two categories with strikingly different properties:
open dots, where the QPC leads are set to fully transmit at least one mode, and
19
closed dots, where all of the leads are set as tunnel barriers. Open dots are impor-
tant laboratories for interference effects, quantum chaos, and the various statistical
ensembles available due to, for example, shape distortions, magnetic fields, and spin-
orbit coupling [44, 45, 46, 47, 48]. All of the experiments in this thesis concern closed
quantum dots, and the remainder of this section will introduce the physics of these
systems.
A closed quantum dot can be represented schematically as an island of charge con-
nected capacitively and via tunnel barriers to conducting reservoirs, and purely ca-
pacitively to any number of gates. For simplicity here I will assume source and drain
reservoirs (with capacitances Cs and Cd and voltages Vs and Vd and a single gate (Cg
and Vg) as shown in Fig. 3.2(a). The Hamiltonian of the dot can then be written as
HQD = EcN(N − 1)
2− Ne
CΣ
(CgVg + CsVs + CdVd) +∑i,σ
Niσεiσ (3.5)
where CΣ is the sum of all gate and lead capacitances and is related to the charg-
ing energy Ec by Ec = e2/CΣ. The first term is the Coulomb energy associated
with putting N electrons onto the island together, making the constant interaction
approximation that every electron feels a Coulomb repulsion energy Ec with every
other electron, regardless of what quantum state either electron occupies. The sec-
ond term in Eq. 3.5 is the Coulomb interaction between electrons in the dot and
the other nearby conductors, where CΣ is the sum of all the individual capacitances.
Again we’ve assumed constant interactions with these conductors independent of the
quantum state of each electron. There are several trivial corrections to the constant
interaction approximation, such as a general decrease in Ec as N increases. Simple
electrostatics imply that Ec should vary somewhere between inversely with area (if
the dot/gate geometry looks like a parallel plate capacitor) and inversely with perime-
20
ter (if the dot is large and coplanar with the gates). Typical values of Ec in GaAs
quantum dots range from several meV in few-electron dots to a few hundred µeV in
large (∼1000 electrons) dots. There are also deviations from the constant interaction
model with more interesting effects, such as Hund’s rule [49], which we will touch on
in Ch. 7.
These first two terms are essentially classical, requiring only that charge be quantized.
The final term in Eq. 3.5 is due to quantum confinement in the electrostatic potential
of the gates. Here we make the approximation that each electron enters a single-
particle energy level, where the i-th level with spin σ has energy εiσ and is occupied
by Niσ = 0 or 1 electrons. The many-particle state is then simply the product of
the occupied single-particle states. The validity of this approximation is the subject
of countless works, both theoretical and experimental, in a wide variety of quantum
systems; Ref. [50] provides an introduction to this question as it concerns 2DEGs,
and shows that the single-particle picture is remarkably successful in dots containing
several hundred electrons. More recent studies of few-electron quantum dots have
shown the single-particle picture to work well enough that a shell structure similar to
atomic energy levels can be observed [25], leading these systems to be called “artificial
atoms.” Later chapters will have more to say about the spectra of energy levels εiσ
in a quantum dot, but for now it is only important that they are arrayed with some
mean energy spacing ∆ between successive levels of the same spin.3 Since the density
of states in 2D is a constant, in large dots where the effect of the gates on the
potential is screened except near the perimeter, ∆ goes as inverse dot area, or simply
∆ = gsEf/N where Ef is the Fermi energy measured relative to the band minimum.
3For the most part this is the relevant spacing, because even when spin degeneracyis broken, states of opposite spin will hybridize with each other only very weakly.
21
Ef is around 7 meV in the 2DEGs used here, so a 0.5 µm2, 1000-electron dot would
have ∆ ≈ 14µeV. The gate potential is not screened in the middle of small dots so
the band minimum rises, decreasing the level spacing. The first excited state of one
of our one-electron dots is typically 1–3 meV above the ground state. In GaAs dots,
almost always ∆ < Ec, although in some other systems (nanotubes, for example) this
is not necessarily the case.
In addition to the Hamiltonian of the dot itself, the source and drain reservoirs (α ∈
S,D) have energies
Hα =∑k,σ
εkNαkσ (3.6)
where k are states in the leads (plane waves if we assume the leads are clean 1D chan-
nels) with spins σ and occupations Nαkσ which usually4 follow a Fermi distribution
N(ε) = 1/(1 + eµα−εkTα ) based on the chemical potential µα = −eVα and temperature
Tα of the reservoir. The reservoirs couple to the dot via terms
HQD−α =∑i,k,σ
tiαkσ(c†iσcαkσ + h.c.) (3.7)
which link creation and annihilation operators c† and c in the dot with those in the
leads via tunneling elements tiαkσ that depend on the particular form of the i-th
wavefunction in the dot (whether, for instance, it has a node in the vicinity of the
tunnel barrier5) and can in principle be functions of energy or spin of electrons in
the leads as well.6 h.c. denotes the Hermitian conjugate. Fermi’s golden rule yields
a transition rate into and out of each level in the the dot based on these tunneling
4See Ref. [51] for an application in which a non-Fermi distribution is created andstudied.
5See Ref. [52] for a description of this variation using random matrix theory, ap-plicable to large dots.
6Refs. [53, 54] use spin-split Landau levels to load particular spins into a dot, andRef. [55] discusses a similar scheme in parallel (Zeeman) field.
22
amplitudes,
Γαiσ = 4πρ(εiσ)|tα|2 (3.8)
where ρ(εiσ) is the density of states in lead α at the energy of the dot level.7 Γ has
units of energy, from which the tunneling time is given by tαiσ = ~/Γαiσ. Γ has
additional significance as the lifetime broadening of the states in the dot.
Once the dot is closed and charge is quantized, we encounter the phenomenon of
Coulomb blockade. At most values of gate voltage only one charge state is allowed, so
no charge may be added to or removed from the dot, and conductance is suppressed.
Consider first the condition Vsd = 0 (for simplicity, choose Vs = Vd = 0). Ignoring for
the moment the quantum confinement term and adding an irrelevant term depending
only on gate voltage, we can rewrite Eq. 3.5 as
Hdot = (Ec/2)(N −Ng)2 (3.9)
whereNg = CgVg/e+1/2. The gate chargeNg increases linearly with gate voltage, but
in the ground state, the real charge N on the dot rounds this to the nearest integer,
as depicted in Fig. 3.2(b). The only values of gate voltage at which conductance is
allowed are where Ng is close to a half-integer n+ 12, at which point electrons may pass
through the dot one at a time, alternating between charge states n and n + 1. This
leads to conductance spikes as a function of gate voltage, with a period ∆Vg = e/Cg
(Fig. 3.2(b), bottom panel).
In total there are five energy scales relevant to transport through a quantum dot,
7ρ here is the absolute density of states, with units of inverse energy, not the usualdensity per unit length. As the length of the lead is increased, ρ increases linearly,while the tunneling elements t vary as the inverse square root of lead length since theyare proportional to individual wavefunction amplitudes. Thus, Γ tends to a constantas lead size diverges.
23
Rs
QD
Vg
Cg
VdVs
CdCs
Rd
A
Vg
G
NgN
a) b)
Ec
∆
kT eVsd
c) d) e)Γ
Figure 3.2: Schematic description of a quantum dot. a) Circuit equivalent of a quan-tum dot, showing voltages and capacitances of the source, drain, and gate, tunneling“resistances” to the source and drain, and the current measurement for calculatingconductance. b) The smooth gate chargeNg, resulting dot chargeN , and conductancespikes characteristic of Coulomb Blockade. c) Five energy scales determine behaviorof a closed quantum dot. A gap Ec opens up between the occupied n-electron states(solid lines) and unoccupied n+ 1-electron states (dotted lines), which are spaced by∆ and tunnel-broadened by Γ, while the leads are characterized by temperature kTand bias eVsd. A transport resonance condition in the quantum regime d) at zerobias, where current flows through only the ground state, and e) at Vsd > ∆, throughground and excited states. The thick solid line represents the newly occupied statein the n + 1-electron ground state. The thin solid and dotted lines represent statesfilled or empty, respectively, in the ground states of both occupancies.
shown on a generic level diagram in Fig. 3.2(c). Two involve the structure of the dot
(Ec and ∆), two concern the leads (kT and eVsd), and one describes the coupling
between them (Γ).8 The first question is what determines whether the dot is open or
8To determine the conductance through the dot it is necessary to know the tunnelrate to each lead, but for delineating regimes of different behavior only their sum isrelevant.
24
closed. The Hamiltonian Eq. 3.5 is only valid if charge in the dot is quantized, but
when the dot is open, charge in the leads flows continuously into the dot to screen
charge excitations. There are three ways I know of to get at this crossover, which
all seem independent until they arrive remarkably at the same answer. First, we
may simply stipulate that the entrance and exit must be tunnel barriers, since this
constrains motion through the barrier that would allow screening. In other words,
Gα < e2/h. A second, more satisfying route involves the energy-time uncertainty
relation, noting that charge may only be quantized if energy can be resolved to better
than Ec. The time that determines energy resolution is the charge relaxation time
RαCΣ = CΣ/Gα, and using ∆E∆t > h again leads directly to Gα < e2/h.
Finally, for a golden-rule approach to be valid and not give way to higher-order
processes (which in turn screen the charge), the energy spectrum in the dot must
be well-resolved, meaning that Γ < ∆. Although it doesn’t explicitly make sense to
talk about a resistance unless the barrier connects two continuums, we can regard
the resistance as a property of the barrier independent of what states it connects,
and calculate it by approaching the continuum limit in the dot. One way to do this
without changing the properties of the dot (i.e. letting ∆ → 0) is to assume a large
bias across the barrier so that tunneling is allowed into n = eV/∆ states. The current
across the barrier is I = edNdt
= enΓ/~ = V e2
~Γ∆
and dividing by voltage gives
G =e2
~Γ
∆. (3.10)
Within a factor of 2π, the condition Γ < ∆ is then equivalent to G < e2/h.
The heirarchy Γ < ∆ < Ec for Coulomb-blockaded dots yields four temperature
ranges, each with its own distinct behavior. The effect of temperature is to broaden
25
the step from filled to empty states in the reservoirs by the Fermi function, with an
energy width proportional to kT . At the highest temperatures, kT > Ec, conductance
is never fully suppressed, because thermal excitations can always access multiple
charge states, and conductance may show weak oscillations with period e/Cg if kT is
only somewhat larger than Ec. This limits large closed dots (Ec ∼ 100µeV) to use
well below 1 K (= 86µeV) and for practical purposes they are best studied in dilution
refrfigerators at below 100 mK. Small dots, where the energy scales are larger, may
be studied in warmer cryostats (3He at 300 mK or 4He at 1 K), however even the
small-dot experiments presented here were performed in a dilution refrigerator for
maximum energy resolution.
The range ∆ < kT < Ec is known as classical Coulomb blockade, because while
transport is fully suppressed away from resonance and on resonance only two charge
states contribute, multiple energy levels are available within each charge state. Levels
in the dot are filled according to a Fermi distribution, just like states in the lead, and
the convolution of these three Fermi functions gives a peak with cosh−2 voltage profile
(i.e. exponential tails) and a full width at half maximum (FWHM) of 4.3kT when
scaled by the lever arm Cg/CΣ relating voltage changes on the gate to energy changes
in the dot.
Cooling further (or using smaller dots) we reach the quantum Coulomb blockade
regime, kT < ∆, where only one energy level is involved in transport. This regime
is divided further into the thermally broadened regime Γ < kT (warmer or more
closed barriers) and the lifetime-broadened regime Γ > kT (cooler or more open
barriers). Thermally broadened peaks still have a cosh−2 voltage profile but now only
two Fermi functions (one for each lead) are convolved and the scaled FWHM drops
26
to 3.5kT . In addition, the peak height decreases as temperature increases, whereas
in classical blockade the peak height is roughly constant with temperature. In the
lifetime-broadened regime, all Fermi functions are out of the picture, and instead the
voltage profile of the peak is Lorentzian (i.e. polynomial tails) characteristic of a
resonance damped by decay to the leads. The quantum Coulomb blockade regime is
accessible in a dilution refrigerator for dots containing several hundred electrons or
less (∆ > 50µeV), which describes all of the dots used in this thesis. What makes
this regime interesting is that charge motion is dependent on the properties of the
individual quantum state being accessed, and many of these properties can be deduced
from measurable quantities like peak position and height, and how these change in
magnetic fields. In addition, as we will see in Ch. 9, this regime allows preparation of
a known quantum state which can then be manipulated to perform more complicated
dynamics studies.
Figure 3.2(d) shows the spectrum of single-particle levels available at the n to n+ 1
electron resonance with zero source-drain bias. In the quantum Coulomb blockade
regime at this condition, current flows through only the thick-lined state, which is
empty in the n-electron ground state and filled in the n+1-electron ground state. The
charging energy gap is now effectively absent from the system: higher energy levels
(dashed lines) are available to accept electrons when the thick state is vacant, and
lower energy levels are available to eject electrons when the thick state is filled without
paying the charging energy cost. At zero bias such processes are forbidden because
the occupation of states in the leads exactly matches the occupation within the dot,
and no filled and empty levels exist at the same energy. But at sufficient source-drain
biases, so long as the thick state remains between the two Fermi levels, these processes
27
are allowed. Current through the dot changes whenever one of these states passes
either the source or drain Fermi level, enabling excited-state spectroscopy of the dot
by measuring current vs. Vg and Vsd [50].
3.3 Double dots
A QPC by itself couples two continuums to each other, and a single quantum dot
couples a continuum to a discrete set of states. We can investigate the coupling
between two discrete spectra by creating two quantum dots tunnel-coupled to each
other. The circuit equivalent of a double dot is shown in Fig. 3.3(a). The key new
features are the mutual capacitance and interdot tunneling, denoted by Cm and Rm,
which parametrize the interaction between the two dots. Likewise, following Ref. [40],
we can build up the double-dot Hamiltonian from single-dot components plus a term
for each form of interaction,
HDQD =Ec1
2N(N − 1)− NEc1 +MEm
e(Cg1Vg1 + CsVs) +
∑i,σ
Niσεiσ
+Ec2
2M(M − 1)− MEc2 +NEm
e(Cg2Vg2 + CdVd) +
∑j,σ
Mjσεjσ
+ EmNM +∑i,j,σ
tijσ(c†iσcjσ + h.c.). (3.11)
Here N and M are the occupations of the left and right dots, and defining C1(2) as
the total capacitance of the left (right) dot, the single dot charging energies are given
by
Ec1(2) =e2
C1(2)
(1− C2
m
C1C2
)−1
(3.12)
28
VV
sg1
sC Rs
C
a
ddC R
g1 C
C Rmm
g2
g2
VVs
QD1 QD2 (n+1, m+1)
(n+1, m)
(n, m+1)
(n, m)
1µ (n+1,m+1)= -|e|V
µ (n+1,m+1)= 01
(n, m) (n+1, m)
(n, m+1)
µ (n+1,m+1)= 02
E
E
E 2|t|n,m+1
n+1,m
∆V
E.S.
G.S.
b
c d
Vg
2
Vg1
A
∆V
(n+1, m+1)
Vg
2
Vg1
Figure 3.3: Double quantum dots. a) Circuit equivalent of a double dot. The doubleboxes represent a capacitor and resistor in parallel. b) At zero source-drain bias,the ground state charge is constant within hexagonal regions in the Vg1–Vg2 plane.Ordered pairs (n,m) denote charge on the left and right dots respectively. Currentcan only flow at the vertices, called triple points, where three charge states are inequillibrium. The upper one (top diagram) is called the hole triple point and thelower one (middle diagram) the electron triple point. When the dots are fairly open,conduction is sometimes seen along the hexagon edges as well, due to cotunnelingprocesses as in the bottom diagram. When considering two states of the same totalcharge, e.g. (n,m+ 1) and (n+ 1,m), the relevant parameter is detuning, or motionaway from equal energy such as the line labeled ∆V . c) At finite source-drain bias,the triple points expand into triangles. Ground-state-to-ground-state transport occursalong the base of each triangle (∆V = 0), and excited states are manifest as lines ofcurrent parallel to the ground state line. d) Due to a finite tunnel coupling t betweenthe dots, two states of the same total charge will hybridize, displaying an avoidedcrossing as a function of detuning. Figure adapted from [40] and [56].
29
and the mutual charging energy (the extra energy needed to add an electron to one
dot due to a single electron on the other) is
Em =e2
Cm
(C1C2
C2m
− 1
)−1
(3.13)
Tunnel elements tijσ connect states i in the left dot with states j in the right dot,
conserving the spin σ. The full Hamiltonian contains additional terms corresponding
the lead and to dot-lead couplings, which are identical to their single-dot counterparts.
The only additional approximations incorporated in Eq. 3.11 beyond those made for
single dots are that cross-capacitances are zero (e.g. from gate 1 to dot 2—this is a
fairly trivial assumption to remove, by using linear combinations of Vg1 and Vg2.) and
that the tunneling elements are small enough that the two dots are really distinct, and
as far as charging effects are concerned electrons are localized on one dot at a time.
As with single dots, we can make the equivalent assertions that electron localization
begins when the RC time is long enough to resolve the charging energies or when
the barrier has a tunneling conductance (Gm < e2/h). We can also guess that strong
mixing of states in the two dots would occur when the tunneling elements tijσ become
equal to the level spacing ∆ of each dot (which we assume are equal for simplicity).
Barrier conductance is perhaps even less meaningful in a double dot than in a single
dot, but we can relate it to the tunneling elements through Eqs. 3.8 (taking ρ = 1/∆)
and 3.10, yielding Gm = 4π e2
h( t
∆)2. Again, to within a constant, all three approaches
give the same condition, t < ∆, for the dots to be well-separated.
As with single dots, we begin by considering only the classical terms in Eq. 3.11,
omitting the last term on each line which is a purely quantum effect, and for the
moment set the source and drain voltages to zero to examine the ground state of the
30
system. With the substitutions Ng = Cg1Vg1/e+K1 and Mg = Cg2Vg2/e+K2 where
K1(2) = 12
1−Em/EC1(2)
1−E2m/EC1EC2
is just a constant offset,9 we can rewrite the classical energy
as
U(N,M) =1
2Ec1(N −Ng)
2 +1
2Ec1(N −Ng)
2 + Em(N −Ng)(M −Mg). (3.14)
Because E2m < Ec1Ec2,
10 (N,M) = (Ng,Mg) is the ground state charge configuration
for every set of gate voltages such that Ng and Mg are integers. In Vg1–Vg2 space,
therefore, the ground states map out a grid. In the limit of Em = 0, the charge
transitions in each dot exactly follow the gridlines of that dot’s own gate irrespective
of the other gate voltage, shown by the dotted lines in Fig. 3.3(b). Current would
pass through the system only at the intersections, where four charge states would
be degenerate. Finite mutual charging is equivalent to an energy term depending
on the total double-dot occupation N + M , and because the lower left and upper
right states at this intersection differ in total occupation by two (one in each dot),
the mutual charging term breaks their degeneracy and splits the intersection point
into two, called triple points because three charge states are degenerate. With this
splitting, each charge configuration is the ground state over a hexagonal region in gate
space, tiling into a honeycomb pattern as indicated by the solid lines in Fig. 3.3(b).
Current flows at both triple points by the same tunneling processes and through the
same states in each dot, however the order of tunneling events is different depending on
whether the doubly-empty or doubly-occupied charge state completes the conduction
9This offset, like the 1/2 offset in the single-dot case, arises from the N(N − 1)factor (rather than N2) due to the quantized charge on the dot. In practice it isirrelevant because we don’t know beforehand what voltage should correspond to azero-electron dot, which is likely why many authors simply use N2 and ignore theoffset.
10This can be shown to always be true, using Eqs. 3.12 and 3.13 and the fact thatC1 and C2 are sums which both include Cm.
31
cycle. Consider a small increase in the Fermi energy of the left lead (i.e. a negative
voltage) such that electrons flow from left to right. At the upper, or “hole” triple
point (top diagram in Fig. 3.3(b)), we can think of the doubly-occupied state as the
initial state, and a single hole enters from the right lead, hops from the right dot
to the left, then escapes to the left lead. At the lower, or “electron” triple point
(middle diagram), the doubly-empty state can be thought of as the initial state, and
an electron enters from the left, hops from the left dot to the right, and escapes to
the right lead. The result is the same, and barring any voltage or mutual charging
dependence of the tunnel rates the current will be the same.
Experimentally, current is often seen along the honeycomb edges as well as at the
triple points. This requires a cotunneling event as shown in the bottom diagram in
Fig. 3.3(b). The left dot is at a charge resonance, so an electron can tunnel from the
left lead to the left dot, then to get to the right lead it may virtually transit through
an energetically forbidden state in the right dot. A similar process can occur on the
short honeycomb edge connecting the two triple points. On this edge an electron
may occupy either dot, but it is energetically forbidden for it to exit or for another
electron to enter. With the existing electron in the right dot, a virtual process can
simultaneously eject that electron to the right lead and bring a new electron from the
left lead into the left dot.
When a large source-drain bias is applied across the double dot, each triple point
expands into a triangle satisfying −|e|V ≥ µ1 ≥ µ2 ≥ 0, as shown in Fig. 3.3(c). This
describes the case of Vd = 0 and Vs = V , where V is negative (higher energy). If V is
positive, the inequalities are reversed. The chemical potentials µ1 and µ2 are simply
the energies to add the last electron to the left or right dot with the other dot occupied
32
as required at that triple point. In the electron triangle, for example, the relevant
chemical potentials are µ1(n + 1,m) = U(n + 1,m) − U(n,m) and µ2(n,m + 1) =
U(n,m + 1) − U(n,m). Within the bias triangles we begin to see the effects of
quantized energy levels in the dots. Fully elastic transport from one ground state to
the other is allowed only along the lower right edge of the triangle, where µ1 = µ2,
and consequently this edge is usually a current maximum. If |eV | > ∆, additional
lines of high current appear parallel to the ground state line due to transport through
excited states, as shown by the lower diagram in Fig. 3.3(c). The remaining area of
the triangle is typically filled in with weaker current due to inelastic transitions from
the left dot to the right dot involving emission of either a phonon or a photon.
The one term in the Hamiltonian Eq. 3.11 we have yet to consider is the tunnel
coupling between the two dots (although in the previous section we implicitly assumed
that some tunneling could occur). This term affects energy levels in the dots only in
the vicinity of the triple points and the short honeycomb segment connecting them.
In this neighborhood we can think of the double dot as a two-level system (the ground
states of (n,m+1) and (n+1,m)) with the bare detuning between them parametrized
by a gate voltage ∆V along a diagonal line as shown in Fig. 3.3(b). The tunnel
coupling then produces an avoided crossing between these levels (Fig. 3.3(d)) during
which the two states hybridize and the ground state passes one electron smoothly
from one dot to the other. The hyperbolic shape of the avoided crossing is reflected
in the triple points, which at large tunnel coupling begin to push apart and stretch
into crescents, smoothly connecting the two longer honeycomb edges coming into each
triple point.
33
3.4 Charge sensing
While anyone studying the 0.7 structure would call QPC traces such as Figs. 3.1(d)–
(g) unacceptibly dirty, for their purpose in this thesis—charge sensing—they are all
just fine. Charge sensing is the use of a conduction path to infer the number and/or
location of nearby charges. The only requirement to be able to do this is that the
conduction path be acutely sensitive to the local electrostatic environment, so that
when an electron moves nearby—into or out of a quantum dot, or from one dot to
another—the conductance of the path changes.
Conceptually, the technique of charge sensing is identical to a field effect transistor—
one might even call charge sensors “single-electron FETs.” There are fundamental
quantum mechanics questions involved in charge sensing, because it is a form of
measurement and therefore exerts a complementary backaction on the system being
measured. For the experiments in the following chapters this backaction will be
negligible in most cases, because the sensors are weakly driven or weakly coupled to
the system being sensed, such that the measurement timescale is much longer than
other electron dynamics.
There is no fundamental limit to the sensitivity of a charge sensor. The location of
a single electron could in principle determine whether a nearby conduction path is
fully opaque or fully transparent, and it has been shown (albeit for fairly low on-state
conductances) that one may alter conductance by more than two orders of magnitude
(see Ref. [57] and Ch. 4). However, most of the charge sensors used in this thesis have
much lower sensitivities, primarily because the sensed electron is screened by a gate
between it and the sensor channel. The key feature of charge sensing relative to
34
transport measurements is that they do not require fast processes within the dots
being measured in order to generate a measureable current. Electrons may move
faster than the measurement time, in which case the sensor will measure an average
property as is done in transport measurements, or they may move more slowly than
the measurement,11 in which case every event may be recorded individually, known as
a single-shot measurement. In some cases (see Ch. 6) the electrons may never visit any
reservoir, so they will generate no current whatsoever, but as long as their positions
change they may be studied via charge sensing. The experiments in the following
chapters explore some of the behavior of charge sensors and their interaction with
single and double quantum dots, then use them as tools to investigate some otherwise
inaccessible energetics and dynamics in double dots.
11Measurements as fast as several µs have been performed in GaAs systems [58, 55].
35
Chapter 4
Coulomb-modified Fano resonancein a one-lead quantum dot
A. C. Johnson, C. M. MarcusDepartment of Physics, Harvard University, Cambridge, Massachusetts 02138
M. P. Hanson, A. C. GossardDepartment of Materials, University of California, Santa Barbara, California 93106
We investigate a tunable Fano interferometer consisting of a quantum dot coupled via
tunneling to a one-dimensional channel. In addition to Fano resonance, the channel
shows strong Coulomb response to the dot, with a single electron modulating channel
conductance by factors of up to 100. Where these effects coexist, lineshapes with up
to four extrema are found. A model of Coulomb-modified Fano resonance is developed
and gives excellent agreement with experiment.1
1This chapter is adapted from Ref. [59]
36
4.1 Introduction
The interplay between interference and interaction, in its many forms, is the central
problem in mesoscopic physics. In bulk systems, screening reduces the strong repul-
sion between electrons to a weak interaction between quasiparticles, but in confined
geometries electron-electron interaction can dominate transport. The Fano effect—an
interference between resonant and non-resonant processes—was first proposed in the
context of atomic physics [60]. More recently, Fano resonances have been investi-
gated in condensed matter systems, including surface impurities [61], quantum dots
[62, 63, 64, 65, 66], and carbon nanotubes [67], and have generated interest as probes
of phase coherence [68, 69] and as possible spin filters [70]. These studies have treated
Fano resonance as purely an interference effect. When the resonant channel is a tun-
neling quantum dot, however, Fano resonance coexists with Coulomb interaction that
appears in single-dot transmission as the Coulomb blockade. In the Fano regime, the
coexistence of Coulomb and interference effects leads to new transport regimes that
to date have not been investigated theoretically or experimentally.
In this chapter, we present measurements of a Fano interferometer consisting of a
quantum dot coupled to a one-dimensional channel. We have independent control of
all couplings defining the resonant and non-resonant processes, allowing us to identify
and investigate several regimes of behavior. With the channel partially transmitting
but not allowing tunneling into the dot, a charge sensing effect is observed whereby
channel conductance responds to the number of charges on the dot [71, 72, 73]. Al-
ternatively, if the dot-channel barrier is lowered, and the conducting path pushed
towards the dot, a regime emerges in which all transmitted electrons pass through
37
1 µm
Source
Drain
Vg
0.8
0.7
0.6
0.5
0.4
0.3
g (
e2/h
)
-1840 -1820 -1800 -1780Vg (mV)
Figure 4.1: Channel conductance data (squares) and fits (curves) vs. gate voltage Vg
in the Fano regime. Bars show fitting ranges. Inset: SEM image of a similar sample,a quantum dot coupled by one lead to a conducting channel.
the dot, and standard Coulomb blockade resonances result. Between these extremes
lies the Coulomb-modified Fano regime, in which resonant tunneling via the dot and
direct channel transmission coincide. Fano resonances appear throughout this regime,
generally in conjunction with charge sensing. We develop a model that combines these
two effects and successfully describes the data. With this model, resonance parame-
ters are extracted and used to evaluate interaction effects.
The device (inset to Fig. 4.1) consists of a 0.5 µm2 quantum dot and a constriction, all
defined by Cr-Au depletion gates on a GaAs/AlGaAs heterostructure grown by MBE.
The two-dimensional electron gas lies 100 nm below the surface, with density 2×1011
cm−2 and mobility 2×105 cm2/Vs. The dot contains ∼ 600 electrons and has level
spacing ∆ ∼ 20 µeV. The experiment was performed in a dilution refrigerator with
38
an electron temperature of 50 mK, in a magnetic field of 0 to 200 mT perpendicular
to the device plane. Conductance was measured using a lock-in amplifier with 10 µV
excitation at 15.7 Hz.
Figure 4.1 shows conductance in the Fano regime as a function of gate voltage Vg
over a range containing seven resonances. A progression of lineshapes is seen, each
comprising a dip and a peak similar to Fano resonance. The non-interacting Fano
lineshape, given by Eq. 4.1 below, is insufficient to explain these data. However, a
model incorporating Coulomb interaction (described below) provides improved fits,
as seen in Fig. 4.1.
4.2 Device behavior
We first discuss qualitatively the limiting regimes with the dot either capacitively or
tunnel coupled to the channel, as well as the intermediate case, the Coulomb-modified
Fano regime. When the channel is tuned to partially transmit one mode, but there
is no conductance between channel and dot, sawtooth patterns such as Fig. 4.2(a)
appear. These can be explained by considering a single effective gate, combining the
effects of the metal gate and the dot charge, that smoothly modulates g, the channel
conductance. Every electron residing on the dot has a gating effect on the channel,
modifying the effective gate voltage by an amount Vs, which we denote the sensing
voltage. Each time a charge is added to the dot (as Vg is made more positive), the
effective gate voltage jumps by Vs, causing g to jump to the value it had at a gate
voltage more negative by Vs. If g is a decreasing (increasing) function of Vg, this
results in an upward (downward) jump. For this device Vs is typically ∼80% of the
39
D
SVg
D
SVg
D
SVg
100
80
60
40
20
0
g (
10
-3e
2/h
)
-1840 -1820 -1800 -1780 -1760Vg (mV)
(c)
0.5
0.4
0.3
0.2
g (
e2/h
)
-1820 -1800 -1780 -1760 -1740Vg (mV)
(b)
0.6
0.5
0.4g
(e
2/h
)
-1880 -1860 -1840 -1820 -1800 -1780 -1760Vg (mV)
(a)
Figure 4.2: Three configurations with a tunnel coupled dot. Drawings indicate tun-neling paths. (a) Pure charge sensing: the dot couples capacitively to the channeland tunnels weakly to a third reservoir. (b) Fano resonance with charge sensing:tunneling between the channel and the dot interferes with direct transmission. (c)Breit-Wigner resonances: the only conducting path is through the dot.
spacing between jumps, indicating that the channel is more sensitive to excess dot
charge than to the gate directly.
The Coulomb-modified Fano regime emerges as tunneling is introduced between the
dot and the channel. When the charge sensing effect is relatively weak, such as in
Fig. 4.1, features clearly resembling non-interacting Fano resonance are observed. In
40
other cases, for example Fig. 4.2(b), the charge-sensing jump is comparable to or
larger than the peak of the Fano resonance. In this regime, combining the charge-
sensing jump with the dip-peak pair of Fano resonance, one resonance can have up
to four extrema.
When direct conductance is made much smaller than conductance through the dot,
the direct path (through the channel) no longer interferes significantly with the reso-
nant path (though the dot). In this limit, the Fano regime crosses over to the familiar
Coulomb blockade regime, yielding simple single resonances as seen in Fig. 4.2(c).
Before analyzing the Fano regime in detail, we turn our attention to features that
can appear as Fano resonance but are actually a result of charge sensing. Figure 4.3
shows the effect of quantized charge in the dot on a resonance in the channel, in the
absence of any tunneling between the two. Channel conductance was measured while
tuning the coupling between the dot and a third reservoir. When the tunnel rate
to the third reservoir is low (Fig. 4.3(a)), one finds smooth segments punctuated by
jumps. The dotted curves are identical forms offset by Vs = 4.0 mV, indicating how
channel resonances would appear for an isolated dot occupied by n = N or n = N+1
electrons. If this family of curves is extended to all n, it overlays every segment of
the data, though the jumps from one curve to the next are irregularly spaced and
change position if the sweep is repeated. We identify these jumps as single tunneling
events, and estimate a tunneling time ~/Γ here of order seconds, based on motion of
the jump point upon repeating sweeps.
As tunneling to the third reservoir is increased (Fig. 4.3(b)), jumps become periodic
and repeatable, but there remains a family of evenly spaced curves onto which all of
41
h/Γ∼sec
Vg
D
S
h/Γ∼µsec
Vg
D
S
h/Γ∼nsec
Vg
D
S40
30
20
10
0
g (
10
-3e
2/h
)
-580 -560 -540 -520Vg (mV)
(c)
40
30
20
10
0
g (
10
-3e
2/h
)
-640 -620 -600 -580Vg (mV)
(b)
0.3
0.2
0.1g
(e
2/h
)
-860 -840 -820 -800 -780 -760 -740Vg (mV)
(a)
Figure 4.3: Charge sensing by a channel resonance. Drawings indicate the tunnelingrate to the extra lead. Dotted curves show how the full resonance would look withdot charge fixed at two consecutive values. (a) With the dot nearly isolated, jumpsbetween these curves are sharp, erratic and unrepeatable. (b) Increase tunneling andthe jumps become periodic and repeatable. (c) With the dot nearly open, the jumpsbroaden and resemble an oscillation superimposed on a single, broad resonance.
the data falls. In this case the channel resonance is much narrower, and adding a
single charge to the dot shifts the channel from directly on to far off resonance. This
single-electron switch has an on/off conductance ratio of 20. Traces with still nar-
rower resonances show on/off ratios >100, as in other recent reports [74, 57]. Notice,
however, the similarity between these asymmetric line shapes and Fano resonances,
42
even though this is pure charge sensing.
Still greater coupling of the dot to the third reservoir yields lifetime-broadened tran-
sitions. This is the case in Fig. 4.3(c), where the dominant feature is a single broad
resonance, modulated by weak charge quantization in the dot. This motivates an im-
portant feature of the model, that the Fano resonance and the charge-sensing jump,
since both result from the same process, have a single width parameter Γ.
4.3 Model for Coulomb-modified Fano resonance
In the single-level transport regime, kT < Γ < ∆, transmission through one discrete
level produces a Breit-Wigner resonance, represented by a complex transmission am-
plitude t = t0/(ε + i) with dimensionless detuning ε = (E − E0)/(Γ/2). Here an
electron of energy E encounters a resonance at E0 with width Γ, and peak transmis-
sion t0, accounting for lead asymmetry. Conductance is proportional to the trans-
mission probability |t|2, giving a Lorentzian lineshape [75]. A continuum channel
can be added to the amplitude (coherently) or probability (incoherently), giving the
non-interacting Fano lineshape,
g(E) = ginc + gcoh|ε+ q|2
ε2 + 1, (4.1)
where gcoh (ginc) is the coherent (incoherent) contribution to the continuum conduc-
tance. The Fano parameter q selects from a symmetric peak (q = ∞), symmetric dip
(q = 0), or a dip to the left (q > 0) or right (q < 0) of a peak. The resonances in
Fig. 4.1 evolve from q = 2.5 on the left to q = 0.6 on the right. In cases allowing
arbitrary phase between the resonant and non-resonant paths, such as the Aharonov-
43
Bohm interferometer, q becomes complex. This is equivalent to increasing ginc. In the
present context, the interfering paths enclose no area, forcing a real q. This eliminates
the ambiguity between ginc and Imq, and constrains the Fano lineshape to maximum
visibility [68].
The Coulomb-modified Fano effect can be modeled by extending Eq. 4.1 to include
nearby resonances and charge sensing effects. We first write g as a sum over initial
occupation numbers of the dot,
g(Vg) =∑
n
gn(Vg)pn(Vg). (4.2)
If the dot contains n electrons, the only resonant processes allowed are the n→ n+1
and n− 1 → n transitions, corresponding to the addition or removal of one electron.
The contribution of the n-electron dot is thus
gn(Vg) = ginc(∼V n) + gcoh(
∼V n)
∣∣∣∣∣1 +q(
∼V n)− i
εn− + i+q(
∼V n)− i
εn+ + i
∣∣∣∣∣2
. (4.3)
The allowed resonances have detunings εn± = (eVgCg/Ctot+En→n±1)/(Γ/2), including
a contribution from Vg (in energy units), with a lever arm given by the ratio of gate
capacitance to total dot capacitance. The allowed resonances add coherently2 to
a direct conductance gcoh. ginc is then added to account for multiple modes in the
channel and explicit decoherence processes. Finally, each term is weighted by pn(Vg) =
[tan−1(εn− − tan−1(εn+)]/π, the zero-temperature probability of occupation n. This
can be derived from the Friedel sum rule, which relates changes δ in transmission
phase to fractional changes in dot occupancy, δ(arg(t)) = πδ(〈N〉) mod π, where t is
the Breit-Wigner transmission or reflection amplitude [77, 78].
2Adding amplitudes is an approximation which inflates the extracted coherencefor poorly-separated resonances. Going beyond this requires choosing a microscopicmodel. See Ref. [76].
44
Charge sensing enters Eq. 4.3 via the effective gate voltage∼V n= Vg − nVs. We
expand dependences on∼V n to first order, q(
∼V n) = q0 + dq
dVg, and similarly for gcoh and
gtot = ginc + gcoh. The slope dgtot
dVggives the charge-sensing sawtooth pattern, where
charges in the dot affect the potential seen by charges in the channel. Nonzero dqdVg
and dgcoh
dVgproduce subtle changes to the lineshape near resonance, and reveal how the
potential seen by a charge in resonance depends self-consistently on its probability of
being in the dot. As nearby resonances are generally similar in lineshape, the model
assumes they obey the same linear expressions when calculating their influence on the
tails of the resonance being fitted. This permits fitting with overlapping ranges as
shown in Fig. 4.1, for more accurate determination of the charge-sensing parameters.
One limitation of this model deserves particular attention. A resonance in the channel
makes conductance away from a dot resonance nonlinear on the scale of Vs, while the
model assumes linearity. Thus, while the model trivially fits Fig. 4.2(a) by setting
gcoh = 0, it cannot account for Fig. 4.3 in this manner. However, the asymmet-
ric lineshapes in Fig. 4.3 give a somewhat plausible fit to the model. In order to
unambiguously identify a Fano resonance, it is necessary that the off-resonant con-
ductance varies linearly, which in turn requires that resonances be well separated so
that background behavior can be isolated.
4.4 Analysis and discussion
Figure 4.4 shows fits to the model and the information this yields. Every resonance
in Fig. 4.4(a) was fit four times, using all permutations of varying or fixing at zero
the parameters dqdVg
and dgcoh
dVg, to investigate whether this unusual sensing is necessary
45
to explain the data. The results for one resonance are shown in Fig. 4.4(b). The first
fit holds both parameters at zero while dgtot
dVg, Vs and the five parameters of a basic
Fano resonance are varied. This reproduces all features qualitatively, but quantitative
agreement is much poorer than with the latter three fits, which are nearly identical
and agree with experiment to almost within the noise.3 We therefore conclude that
at least one of q and gcoh is subject to charge sensing, implying that the potential felt
by the charge in resonance is modified by its own probability to be in the dot.
Finally, we consider parameter correlations among subsequent resonances. Figures
4.4(c)–(e) show gtot and gcoh, Γ, and q extracted from each fit to each resonance
in Fig. 4.4(a). Most trends are consistent with general arguments about tunneling
wave functions: as gtot increases, indicating a lower channel potential, Γ increases
due to a lower tunnel barrier, and q decreases to keep peak conductance, given by
ginc + gcoh(q2 + 1), relatively constant. On top of this there appear to be fluctuations
in gcoh, Γ, and q which are expected due to subsequent dot wave functions having
different amplitudes near the barrier.
Two additional features stand out in the data. First, in many instances the fractional
coherence gcoh/gtot is roughly constant for several peaks then jumps abruptly to a
different level for subsequent peaks, as occurs in Fig. 4.4 at Vg = 1700mV, while gtot
and other parameters evolve smoothly. Observed fractional coherence spans a range
from < 10% to > 50%,4 likely due to multiple weakly transmitting modes in the
3In some cases all four fits are noticeably different, but a near-degeneracy is typical.Therefore, the fit is unreliable when both parameters are varied, hence the erraticbehavior and large error bars on the parameters in red.
4This is a conservative bound from the ratio of minimum conductance to gtot.Extracted gcoh/gtot reaches 90%, because charge sensing raises the minimum even ifginc = 0.
46
0.2
0.1
0.0
g (
e2/h
)
-1760 -1720 -1680Vg (mV)
3
2
1
Γ (
mV
)
0.2
0.1
0.0
g (
e2/h
)
4
2
0
q
gtot
gcoh
(a)
(c)
(d)
(e)
Hold dq/dVg
Hold dgcoh/dVg
Vary bothHold both
2
0
-2
q
-1940 -1900 -1860Vg (mV)
1.0
0.8
0.6
0.4
g (
e2/h
)
(f)
(g)
0.15
0.10
0.05
g (
e2/h
)
-1734 -1732 -1730Vg (mV)
(b)
Experiment Fits
Figure 4.4: (a) Experimental data with twelve Fano resonances. In (b) we show oneresonance and its four fits. Of the parameters dq/dVg and dgcoh/dVg, the fits fix atzero both (black), one (green and blue), or neither (red). All four fits are shown,but the latter three are indistinguishable. In (c)-(e) we plot, using the same colorsto denote fitting method, the parameters gtot and gcoh, Γ, and q, from the fits to (a).Panel (f) shows data exhibiting reversals of the sign of q, with extracted q valuesshown in (g).
47
channel which couple differently to the dot. A jump in gcoh/gtot may reflect abrupt
rearrangement of dot wave functions, changing its coupling to each channel mode
while total coupling, measured by Γ, is nearly unchanged.
Second, changes to the sign of q are present but infrequent. An example is the data in
Fig. 4.4(f) and (g), where q flips twice in the observed range. Previous observations of
mesoscopic Fano resonance in transmission [62, 63, 64, 65], including measurements of
a dot in an Aharonov-Bohm ring [79, 80] showed a constant sign of q, sparking intense
debate on why no reversals were seen when simple theory predicts that consecutive
peaks always change sign [78, 81, 82]. In the present geometry, however, as with
recent work on Fano resonance in reflection [66], a constant sign is expected because
the dot meets the channel in only one lead, so there is no freedom of relative sign
between two matrix elements to reverse the phase. Why then are different signs of
q observed here at all? One possibility is that the scattering phase in the channel
changes by π, as if it too passed through a resonance [83]. This requires q to pass
smoothly through zero at a maximum of gtot, which is consistent with the observation
that q is mostly positive in some regions and negative in others, but cannot explain
the flips in Fig. 4.4(f). A more likely explanation for this data is that the source
and drain leads couple to different areas in the dot, due to the spatial extent of
the tunnel barrier. With appropriate nodes in several wave functions, the source
and drain couplings have opposite signs and q reverses sign for several resonances,
exactly as observed in Fig. 4.4(f). In short, both coherence jumps and Fano parameter
flips can be explained by imperfect one-dimensionality of the channel and the tunnel
barrier. In principle, these effects could be used to study wave function properties
not accessible via transmission.
48
We thank A. A. Clerk, W. Hofstetter and B. I. Halperin for useful discussion and
N. J. Craig for experimental contributions. This work was supported in part by the
ARO under DAAD19-99-1-0215 and the Harvard NSF-NSEC (PHY-0117795). ACJ
acknowledges support from the NSF Graduate Research Fellowship Program.
49
Chapter 5
Differential charge sensing andcharge delocalization in a tunabledouble quantum dot
L. DiCarlo, H. J. Lynch, A. C. Johnson, L. I. Childress, K. Crockett, C. M. MarcusDepartment of Physics, Harvard University, Cambridge, Massachusetts
M. P. Hanson, A. C. GossardDepartment of Materials, University of California, Santa Barbara, California
We report measurements of a tunable double quantum dot, operating in the quantum
regime, with integrated local charge sensors. The spatial resolution of the sensors is
sufficient to allow the charge distribution within the double dot system to be resolved
at fixed total charge. We use this readout scheme to investigate charge delocalization
as a function of temperature and strength of tunnel coupling, showing that local
charge sensing allows an accurate determination of interdot tunnel coupling in the
absence of transport.1
1This chapter is adapted from Ref. [84]
50
5.1 Introduction
Coupled semiconductor quantum dots have proved a fertile ground for exploring quan-
tum states of electronic charge and spin. These “artifical molecules” are a scalable
technology with possible applications in information processing. New kinds of classical
computation may arise from quantum dots configured as single electron switches [74]
or as building blocks for cellular automata [85]. Ultimately, coupled quantum dots
may provide a quantum computing platform where the charge states and/or spins
of electrons play a vital role [86]. Charge-state superpositions may be probed using
tunnel-coupled quantum dots, which provide a tunable two-level system whose two
key parameters, the bare detuning ε and tunnel coupling t between two electronic
charge states [40], can be controlled electrically.
In this chapter, we investigate experimentally a quantum two-level system, realized
as left/right charge states in a gate-defined GaAs double quantum dot, using local
electrostatic sensing (see Fig. 5.1). In the absence of tunneling, the states of the
two-level system are denoted (M + 1, N) and (M,N + 1), where the pair of integers
refers to the number of electrons on the left and right dots. For these two states,
the total electron number is fixed, with a single excess charge moving from one dot
to the other as a function of gate voltages. When the dots are tunnel coupled, the
excess charge becomes delocalized and the right/left states hybridize into symmetric
and antisymmetric states.
Local charge sensing is accomplished using integrated quantum point contacts (QPCs)
positioned at opposite sides of the double dot. We present a model for charge sens-
ing in a tunnel-coupled two-level system, and find excellent agreement with experi-
51
gls grs
gdd
12345
9 10 11 12 13
16
15
148
7
6
1µm
Figure 5.1: SEM micrograph of a device similar to the measured device, consisting ofa double quantum dot with quantum point contact charge sensors formed by gates 8/9(13/14) adjacent to the left (right) dot. Series conductance gdd through the doubledot was measured simultaneously with conductances gls and grs through the left andright sensors.
ment. The model allows the sensing signals to be calibrated using temperature de-
pendence and measurements of various capacitances. For significant tunnel coupling,
0.5kBTe . t ∆ (Te is electron temperature, ∆ is the single-particle level spacing
of the individual dots), the tunnel coupling t can be extracted quantitatively from
the charge sensing signal, providing an improved method for measuring tunneling in
quantum dot two-level systems compared to transport methods [40].
Charge sensing using a QPC was first demonstrated in Ref. [71], and has been used
previously to investigate charge delocalization in a single dot strongly coupled to a
lead in the classical regime [73], and as a means of placing bounds on decoherence in
an isolated double quantum dot [85]. The back-action of a QPC sensor, leading to
phase decoherence, has been investigated experimentally [87] and theoretically [88].
Charge sensing with sufficient spatial resolution to detect charge distributions within
a double dot has been demonstrated in a metallic system [72, 89]. However, in metallic
52
systems the interdot tunnel coupling cannot be tuned, making the crossover to charge
delocalization difficult to investigate. Recently, high-bandwidth charge sensing using
a metallic single-electron transistor [90], allowing individual charging events to be
counted, has been demonstrated [58]. Recent measurements of gate-defined few-
electron GaAs double dots [91] have demonstrated dual-QPC charge sensing down to
N,M = 0, 1, 2..., but did not focus on sensing at fixed electron number, or on charge
delocalization. The present experiment uses larger dots, containing ∼ 200 electrons
each (though still with temperature less than level spacing, see below).
The device we investigate, a double quantum dot with adjacent charge sensors, is
formed by sixteen electrostatic gates on the surface of a GaAs/Al0.3Ga0.7As het-
erostructure grown by molecular beam epitaxy (see Fig. 5.1). The two-dimensional
electron gas layer, 100 nm below the surface, has an electron density of 2× 1011 cm−2
and mobility 2× 105 cm2/Vs. Gates 3/11 control the interdot tunnel coupling while
gates 1/2 and 9/10 control coupling to electron reservoirs. In this measurement, the
left and right sensors were QPCs defined by gates 8/9 and 13/14, respectively; gates
6, 7, 15, and 16 were not energized. Gaps between gates 5/9 and 1/13 were fully
depleted, allowing only capacitive coupling between the double dot and the sensors.
Series conductance, gdd, through the double dot was measured using standard lock-
in techniques with a voltage bias of 5µV at 87 Hz. Simultaneously, conductances
through the left and right QPC sensors, gls and grs, were measured in a current bias
configuration using separate lock-in amplifiers with 0.5 nA excitation at 137 and 187
Hz. Throughout the experiment, QPC sensor conductances were set to values in the
0.1 to 0.4 e2/h range by adjusting the voltage on gates 8 and 14.
53
-418
-422
-426
-430V
01)
Vm(
-582-586-590-594V2 (mV)
0 0.01 0.02gdd (e2 / h)
-418
-422
-426
-430
V01
)V
m(
-582-586-590-594V2 (mV)
-3 0 3δgls (10-3 e2 / h)
-582-586-590-594
-10 -5 0 5 10δgrs (10-3 e2 / h)
(a)
(M,N)
(M+1,N)
(M-1,N)
(M,N+1)
(M+1,N+1)
(M-1,N+1)
(M,N-1)
(M+1,N-1)
(M-1,N-1)
(c)(b)
Figure 5.2: a) Double dot conductance, gdd, as a function of gate voltages V2 and V10.White lines indicate the honeycomb pattern. Within each honeycomb cell, electronnumber on each dot is well defined, with M(N) referring to electron number in theleft (right) dot. b, c) Simultaneously measured sensing signals from left (b) and right(c) QPCs. δgls (δgrs) are QPC conductances after subtracting a best-fit plane. Seetext for details. The horizontal pattern in (b) and vertical pattern in (c) demonstratethat each sensor is predominantly sensitive to the charge on the dot it borders.
The device was cooled in a dilution refrigerator with base temperature T ∼ 30 mK.
Electron temperature Te at base was ∼ 100 mK, measured using Coulomb blockade
peak widths with a single dot formed. Single-particle level spacing ∆ ∼ 80µeV for
the individual dots was also measured in a single-dot configuration using differential
54
conductance measurements at finite drain-source bias. Single-dot charging energies,
EC = e2/Co ∼ 500µeV for both dots (giving dot capacitances Co ∼ 320 aF), were
extracted from the height in bias of Coulomb blockade diamonds [92].
5.2 Charge sensing honeycombs
Figure 5.2(a) shows gdd as a function of gate voltages V2 and V10, exhibiting the
familiar ‘honeycomb’ pattern of series conductance through tunnel-coupled quantum
dots [93, 94, 95]. Conductance peaks at the honeycomb vertices, the so-called triple
points, result from simultaneous alignment of energy levels in the two dots with
the chemical potential of the leads. Although conductance can be finite along the
honeycomb edges as a result of cotunneling, here it is suppressed by keeping the dots
weakly coupled to the leads. Inside a honeycomb, electron number in each dot is well
defined as a result of Coulomb blockade. Increasing V10 (V2) at fixed V2 (V10) raises
the electron number in the left (right) dot one by one.
Figures 5.2(b) and (c) show left and right QPC sensor signals measured simultane-
ously with gdd. The sensor data plotted are δgls(rs), the left (right) QPC conductances
after subtracting a best-fit plane (fit to the central hexagon) to remove the background
slope due to cross-coupling of the plunger gates (gates 2 and 10) to the QPCs. The
left sensor shows conductance steps of size ∼ 3×10−3 e2/h along the (more horizontal)
honeycomb edges where the electron number on the left dot changes by one (solid
lines in Fig. 5.2(b)); the right sensor shows conductance steps of size ∼ 1× 10−2 e2/h
along the (more vertical) honeycomb edges where the electron number of the right
dot changes by one (solid lines in Fig. 5.2(c)). Both detectors show a conductance
55
step, one upward and the other downward, along the ∼ 45-degree diagonal segments
connecting nearest triple points. It is along this shorter segment that the total elec-
tron number is fixed; crossing the line marks the transition from (M + 1, N) to
(M,N +1). Overall, we see that the transfer of one electron between one dot and the
leads is detected principally by the sensor nearest to that dot, while the transfer of
one electron between the dots is detected by both sensors, as an upward step in one
and a downward step in the other, as expected.
5.3 Temperature and tunnel coupling
Focusing on interdot transitions at fixed total charge, i.e., transitions from (M+1, N)
to (M,N + 1), we present charge-sensing data taken along the “detuning” diagonal
by controlling gates V2 and V10, shown as a red diagonal line between the triple points
in Fig. 5.3(a). Raw data (no background subtracted) for the two sensors are shown
in Fig. 5.3(b). The transfer of the excess charge from left dot to right dot causes
a conductance step on both QPCs, clearly discernable from the background slope
caused by coupling of gates 2 and 10 to the QPCs.
Also shown in Fig. 5.3(b) are fits to the raw sensor data based on a model of local
sensing of an isolated two-level system in thermal equilibrium, which we now describe.
Varying V2 and V10 along the red diagonal changes the electrostatic energy difference,
or bare detuning ε, between (M + 1, N) and (M,N + 1). The lever arm relating
gate voltage to detuning is set by the slope of the swept diagonal and by several
dot capacitances, and can be calibrated experimentally as described below. When
the tunnel coupling t mixing these two states is small compared to the single-particle
56
-449
-447
V01
)V
m( -583 -581V2 (mV)
0.280
0.290
0.300
gsl
e( 2
)h /
-584 -583 -582 -581
V2 (mV) along diagonal
0.199
0.201
0.203
gsr
e(
2)h
/
(b)
1.0
0.8
0.6
0.4
0.2
0.0
⟨m⟩
M -
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5∆V2 (mV) along diagonal
1.0
0.8
0.6
0.4
0.2
0.0
⟨n ⟩N
-
Te (fit) 102 mK 196 mK 315 mK
(c)
(M+1,N+1)
(M,N+1)
(M+1,N)
(M,N)
(a)
Figure 5.3: a) Double dot conductance gdd as a function of gate voltages V2 and V10 inthe vicinity of a triple point. Same color scale as in Fig. 5.1(a). The detuning diagonal(red line) indicates the fixed-charge transition between (M + 1, N) and (M,N + 1).b) Left and right QPC conductance with no background subtraction (blue points),along the detuning diagonal, with fits to the two-level model, Eq. (2) (black curves).See text for fit details. c) Excess charge (in units of e) in the left and right dot, atT = 30 mK (blue), 200 mK (green) and 315 mK (red). Corresponding values of Te
extracted from the fits (solid curves) are 102, 196 and 315 mK.
level spacings for the individual dots, we can consider a two-level system whose ground
and excited states, separated by an energy Ω =√ε2 + 4t2, consist of superpositions
of (M + 1, N) and (M,N + 1) [96]. The probability of finding the excess charge on
the left dot while in the ground (excited) state is 12(1 ∓ ε/Ω). The excited state is
populated at finite temperature, with an average occupation 1/(1 + exp(−Ω/kBTe)).
The average excess charge (in units of e) on the left and right dots is thus: 〈m〉 −M
〈n〉 −N
=1
2
(1∓ ε
Ωtanh
(Ω
2kBTe
)). (5.1)
57
Our model assumes that each sensor responds linearly to the average excess charge
on each dot, but more sensitively to that on the nearest dot as demonstrated experi-
mentally in Fig. 5.2. The resulting model for sensor conductance is:
gls(rs) = gol(or) ± δgl(r)ε
Ωtanh
(Ω
2kBTe
)+∂gl(r)
∂εε. (5.2)
The first term on the right is the background conductance of the QPC, the second
term represents the linear response to average excess charge, and the third represents
direct coupling of the swept gates to the QPC. As shown in Fig. 5.3(b), our model
gives very good fits to the data. For each trace (left and right sensors), fit parameters
are gol(or), δgl(r),∂gl(r)
∂ε, and Te. In these data, the tunnel coupling is weak, and we may
set t = 0.
Figure 5.3(c) shows the effect of increasing electron temperature on the transition
width. Here, vertical axes show excess charge extracted from fits to QPC sensor con-
ductance data. Sweeps along the red diagonal were taken at refrigerator temperatures
of 30 mK (blue), 200 mK (green) and 315 mK (red). We use the 315 mK (red) data
to extract the lever arm relating voltage along the red diagonal (see Fig. 5.3(a)) to
detuning ε. At this temperature, electrons are well thermalized to the refrigerator,
and thus Te ≈ T . The width of the sensing transition at this highest temperature lets
us extract the lever arm, which we then use to estimate the electron temperature for
the blue (green) data, getting Te = 102(196) mK.
We next investigate the dependence of the sensing transition on interdot tunneling
in the regime of strong tunneling, t & kBTe. Figure 5.4 shows the left QPC sensing
signal, again in units of excess charge, along the detuning diagonal crossing a different
pair of triple points, at base temperature and for various voltages on the coupling
58
gate 11. For the weakest interdot tunneling shown (V11 = −1096 mV), the transition
was thermally broadened, i.e., consistent with t = 0 in Eqs. 5.1 and 5.2, and did
not become narrower when V11 was made more negative. On the other hand, when
V11 was made less negative, the transition widened as the tunneling between dots
increased. Taking Te = 102 mK for all data and calibrating voltage along the detuning
diagonal by setting t = 0 for the V11 = −1096 mV trace allows tunnel couplings t
to be extracted from fits to our model of the other tunnel-broadened traces. We
find t = 10µeV (2.4 GHz) (green trace), t = 16µeV (3.9 GHz) (turquoise trace), and
t = 22µeV (5.3 GHz) (purple trace). Again, fits to the two-level model are quite
good, as seen in Fig. 5.4.
Finally, we compare tunnel coupling values extracted from charge sensing to values
found using a transport-based method that takes advantage of the t dependence of
the splitting of triple points (honeycomb vertices) [97, 40]. In the weak tunneling
regime, t ∆, the splitting of triple points along the line separating isocharge
regions (M +1, N) and (M,N +1) has two components in the plane of gate voltages,
denoted here δV10 and δV2. The lower and upper triple points are found where
the lowest energy M + N + 1 state (the delocalized antisymmetric state) becomes
degenerate with the charge states (M,N) and (M +1, N +1), respectively. Using the
electrostatic model in Ref. [40], we can show that δV10(2) are related to various dot
capacitances and t by
δV10(2) =|e|
Cg10(g2)
(Cm
Co + Cm
+ 2tCo − Cm
e2
). (5.3)
Here, Cg10(g2) is the capacitance from gate 10 (2) to the left (right) dot, Co is the
self-capacitance of each dot, and Cm is the interdot mutual capacitance. All these
capacitances must be known to allow extraction of t from δV10(2). Gate capacitances
59
1.0
0.8
0.6
0.4
0.2
0.0
⟨m⟩
- M
-2 -1 0 1 2∆V2 (mV) along diagonal
(M,N+1)
(M,N+1)(M+1,N)
(M+1,N)
2t
-1096 mV -1080 mV -1074 mV -1070 mV
V11
40
30
20
10
0
t (µ
eV)
-1090 -1080 -1070V11 (mV)
t (sensing)t (transport)
Figure 5.4: Excess charge on the left dot, extracted from left QPC conductance data,along a detuning diagonal (crossing different triple points from those in Fig. 5.3) atbase temperature and several settings of the coupling gate 11. The temperature-broadened curve (red) widens as V11 is made less negative, increasing the tunnelcoupling, t. See text for details of fits (solid curves). Top right inset: comparison of tvalues extracted from sensing (circles) and transport (triangles) measurements, as afunction of V11. Colored circles correspond to the transitions shown in the main graph.Lower left inset: Schematic energy diagram of the two-level system model, showingground and excited states as a function of detuning ε, with splitting (anticrossing) of2t at ε = 0.
Cg10(g2) are estimated from honeycomb periods along respective gate voltage axes,
∆V10(2) ∼ |e|/Cg10(g2) ∼ 6.8 mV. Self-capacitances Co can be obtained from dou-
ble dot transport measurements at finite bias [40]. However, lacking that data, we
estimate Co from single-dot measurements of Coulomb diamonds [92]. Mutual capaci-
tance Cm is extracted from the dimensionless splitting δV10(2)/∆V10(2) ∼ Cm
Co+Cm∼ 0.2,
measured at the lowest tunnel coupling setting.
Tunnel coupling values as a function of voltage on gate 11, extracted both from
charge sensing and triple-point separation, are compared in the inset of Fig. 5.4.
The two approaches are in good agreement, with the charge-sensing approach giving
60
significantly smaller uncertainty for t & 0.5kBTe. The two main sources of error in
the sensing approach are uncertainty in the fits (dominant at low t) and uncertainty
in the lever arm due to a conservative 10 percent uncertainty in Te at base. Error
bars in the transport method are set by the smearing and deformation of triple points
as a result of finite interdot coupling and cotunneling. We note that besides being
more sensitive, the charge-sensing method for measuring t works when the double dot
is fully decoupled from its leads. Like the transport method, however, the sensing
approach is valid for t ∆ (this condition may not be amply satisfied for the highest
values of V11).
In conclusion, we have demonstrated differential charge sensing in a double quan-
tum dot using paired quantum point contact charge sensors. States (M + 1, N) and
(M,N + 1), with fixed total charge, are readily resolved by the sensors, and serve
as a two-level system with a splitting of left/right states controlled by gate-defined
tunneling. A model of local charge sensing of a thermally occupied two-level system
agrees well with the data. Finally, the width of the (M + 1, N) → (M,N + 1) transi-
tion measured with this sensing technique can be used to extract the tunnel coupling
with high accuracy in the range 0.5kBTe . t ∆.
We thank M. D. Lukin, B. I. Halperin and W. van der Wiel for discussions, and
N. J. Craig for experimental assistance. We acknowledge support by the ARO un-
der DAAD19-02-1-0070, DARPA under the QuIST program, the NSF under DMR-
0072777 and the Harvard NSEC, Lucent Technologies (HJL), and the Hertz Founda-
tion (LIC).
61
Chapter 6
Charge sensing of excited states inan isolated double quantum dot
A. C. Johnson, C. M. MarcusDepartment of Physics, Harvard University, Cambridge, Massachusetts 02138
M. P. Hanson, A. C. GossardDepartment of Materials, University of California, Santa Barbara, California 93106
Pulsed electrostatic gating combined with capacitive charge sensing is used to per-
form excited state spectroscopy of an electrically isolated double-quantum-dot system.
The tunneling rate of a single charge moving between the two dots is affected by the
alignment of quantized energy levels; measured tunneling probabilities thereby re-
veal spectral features. Two pulse sequences are investigated, one of which, termed
latched detection, allows measurement of a single tunneling event without repetition.
Both provide excited-state spectroscopy without electrical contact to the double-dot
system.1
1This chapter is adapted from Ref. [98]
62
6.1 Introduction
Electrically controllable discrete quantum states found in quantum dot systems are
efficient laboratories for the study of quantum coherence [99, 100], as well as a poten-
tial basis for quantum computation [101, 102]. Measuring the spectrum and dynamics
of quantum dots requires coupling to a macroscopic measurement apparatus, which in
turn may act to reduce coherence [103, 87, 104, 105, 106]. Excited state spectroscopy
of single [107, 108] and double [109, 110, 40] quantum dots has typically been per-
formed using nonlinear transport, requiring tunnel coupling of the device to electron
reservoirs [40]. This coupling perturbs the quantum states and may increase deco-
herence and heat the device, particularly at the large biases needed for spectroscopy
far from the Fermi surface.
An alternative approach that we investigate in this chapter is to use capacitive charge
sensing [71, 72, 73, 91, 84] combined with pulsed gate voltages that provide an ex-
citation window [111]. Charge sensing has recently been used to probe excited-state
spectra in a few-electron quantum dot coupled to one reservoir [112]. We investigate
pulse/sense spectroscopy in an electrically isolated double quantum dot, where a sin-
gle charge, moving between the two dots, is used to probe excited states. Local charge
sensing by a quantum point contact (QPC) near one of the dots provides readout.
The pulse/sense method operates as follows: A reset pulse on two gates simultane-
ously opens the coupling between the dots and tilts the potential, putting the excess
charge on a selected dot. When the pulse is removed, each dot separately is in its
ground state, but the double dot system is out of equilibrium. The excess charge is
given a finite time to tunnel to the other dot (the probe time). The probability of
63
tunneling depends sensitively on the alignment of ground and excited state levels in
the two dots. Whether or not the charge tunnels during the probe time is detected
by a QPC sensor.
Two gate sequences are investigated. In the first, a short reset pulse is followed by a
relatively long probe interval during which a low tunneling rate gives a moderate total
probability for tunneling. By cycling the reset/probe steps, the QPC measures the
average charge configuration, dominated by the probe step. This allows fine energy
resolution, as the probe process is insensitive to both thermal effects and experimental
difficulties associated with short pulses. The second sequence uses two pulses: the first
resets the system; the second allows weak tunneling between dots. The second pulse
is followed by an arbitrarily long interval (microseconds to hours, in principle) with
the interdot barrier closed, so that each dot is completely isolated, with fixed charge.
This sequence we term latched detection, because the measurement occurs after the
double dot is latched into a final state, separating in time the charge dynamics and
the measurement. If needed, the measurement could be turned off during the probe
pulse, though here we make the measurement time much longer than the probe time
and measure weakly, so that the total back-action of the measurement on the system
during the probe pulse is negligible. While here we use latched detection only for
excited-state spectroscopy, we emphasize that its usefulness is much more general.
The device (Fig. 6.1(a)), defined by e-beam patterned Cr-Au depletion gates on a
GaAs/ AlGaAs heterostructure grown by MBE, comprises two tunnel-coupled quan-
tum dots of lithographic area 0.25 µm2 each and two independent charge-sensing
channels, one beside each central dot. The two-dimensional electron gas lies 100 nm
below the surface, with bulk density 2×1011 cm−2 and mobility 2×105 cm2/Vs. Each
64
12345
6
7
8
9 10 12 13
14
15
16
1 µm
11
-430
-426
-422
-418
V10
(m
V)
-594 -590 -586 -582V2 (mV)
-3 0 3gls (10
-3e
2/h)
-540
-530
-520
-510V
10 (
mV
)
-700 -690 -680 -670V2 (mV)
-4 0 4gls (10
-3e
2/h)
(a)
(b) (c)
gdd
grsgls
Figure 6.1: (a) Scanning electron micrograph of a device identical in design to theone measured, consisting of a double quantum dot with a charge sensor on eitherside. High-bandwidth coaxial lines are attached to gates 11 and 12, dc lines to theother gates. (b) Left sensor signal as a function of gate voltages V2 and V10 revealshexagonal charge stability regions (one outlined with dashed lines) when the doubledot is tunnel-coupled to leads. (c) Sensor signal with the double dot isolated fromthe leads. Only interdot transitions remain. In all pulse/sense experiments presented,the double dot is isolated (as in c) and V2 and V10 are swept along a diagonal crossingthese transitions (solid white line). A plane is subtracted in (b) and (c) to compensatedirect coupling of gates 2 and 10 to the sensor.
dot contains ∼150 electrons and has a single-particle level spacing ∆ ∼ 100µeV (es-
timated from effective device area) and charging energy Ec ∼ 600–700 µeV. Left
65
and right sensor conductances gls and grs were measured using lock-in amplifiers with
2 nA current biases at 137 and 187 Hz; double-dot conductance gdd was measured
using a third lock-in amplifier with a 5 µV voltage bias at 87 Hz, although during
pulse/sense measurement the double dot was fully isolated and the gdd and grs cir-
cuits grounded. The charge sensors were configured as QPCs by grounding gates 6,
7, 15, and 16, and were isolated from the double dot by strongly depleting gates 5,
9, 13, and 1. Measurements were carried out in a dilution refrigerator with electron
temperature ∼100 mK.
6.2 Tunnel-coupled and isolated double dots
Figure 6.1(b) shows the left sensor signal as a function of gate voltages V2 and V10
with the device tunnel-coupled to both leads. Here and in subsequent plots, a plane
has been subtracted to level the central plateau to compensate for capacitive coupling
between the gates and the sensor. In this regime, a honeycomb pattern characteristic
of double-dot transport [40] is seen as a set of hexagonal plateaus in the left sensor
conductance, with horizontally oriented steps of ∼ 3 × 10−3 e2/h corresponding to
changes in the number of electrons in the left dot, controlled by V10, and smaller
vertically oriented steps marking changes in the right dot, controlled by V2. Steps at
the short upper left and lower right segments of each hexagon reflect movement of an
electron from one dot to the other, with total number fixed. Here, an increase in gls
marks an electron moving away from the left sensor, or an increase in the number of
electrons in the right dot.
Transport through the double dot, gdd, occurs only at the honeycomb vertices [84]. As
66
tunneling to the leads is reduced by making V2 and V10 more negative, the honeycomb
sensing pattern persists after gdd has become immeasurably small, but the landscape
changes dramatically as in Fig. 6.1(c) when the tunneling time between the double dot
and the leads diverges. Here, steps follow diagonal lines of constant energy difference
between the dots because only transitions from one dot to the other are allowed.
6.3 Single-pulse technique
Pulse/sense measurements were carried out in this isolated configuration, with the
energy difference between the dots controlled by simultaneously varying V2 and V10
along diagonals (shown, for example, by the white line in Fig. 6.1(c)), and the interdot
barrier controlled by gate voltage V3. Fast control of the same two parameters was
achieved using two synchronized Agilent 33250 arbitrary waveform generators, with
rise times of ∼ 5 ns, connected to gates 11 and 12 via semirigid coaxial lines and
low-temperature bias tees. To compensate a slight cross coupling of gates 11 and 12,
the pulse generators produced linear combinations of pulses, denoted VB (affecting
the barrier, mainly V11) and VE (affecting the energy difference, mainly V12).
The single-pulse/probe sequence is shown schematically in the inset of Fig. 6.2.
Square pulses of length treset = 100 ns are applied every 20 µs, with VB = 100 mV
opening the tunnel barrier while the pulse energy shift VE is varied. The double dot
relaxes to its overall ground state during the reset pulse if treset is much longer than
the elastic and inelastic tunneling times τel and τin while the barrier is pulsed open,
and also longer than the energy relaxation time within the dots. For the next 19.9
µs, the barrier is nearly closed, such that τel < tprobe < τin, and the energy levels are
67
A B C D E
Reset
Probe
1.0
0.5
0.0
⟨n⟩-N
-944 -940 -936 -932V2 (mV) along diagonal
20
10
0
g ls
(10
-3e2 /h
)
AB
C
D
E
time
treset
tprobeVB
(mV
)
0
0
VE
100
Figure 6.2: Single-pulse technique. Time-averaged conductance of the left sensor asa function of V2 along diagonal (see Fig. 6.1(c)) with pulses applied. Inset: Pulseson gates 11 and 12 (parameterized by VB and VE, controlling interdot barrier andrelative energy, see text), followed by a long interval of weak tunneling. A linear fitto the left plateau is subtracted. Right axis shows the average right-dot occupation〈n〉 − N . Points A–E mark features used to infer the excited state spectrum, withschematic interpretations shown below the graph.
returned to their values before the pulse. If elastic tunneling is allowed, the electron
will likely tunnel and then have ample time to relax to its ground state. Without
elastic tunneling, the electron will likely remain where the reset pulse put it.
Figure 6.2 shows the sensing signal during a simultaneous (diagonal, see Fig. 6.1(c))
sweep of V2 and V10, with five points (A to E) labeling key features. The data were
68
taken with VE negative, so that the pulse tilts the ground state toward the left dot.
At point A, the energy difference in both the reset and probe states favors the excess
charge occupying the left dot, and a flat sensing signal corresponding to a time-
averaged right-dot occupation 〈n〉 = N is observed. At the opposite extreme (E),
the energy difference in both cases favors the right dot, giving another a flat signal,
〈n〉 = N + 1. At point B, the ground states of the dots are degenerate in the probe
state (except for small tunnel splitting). This degeneracy appears in the data as a
small peak in right-dot occupation, often barely visible above the noise, presumably
because tunneling in this case is reversible; the electron does not relax once it enters
the right dot, so it is free to return to the left. There is a much larger peak at C. Here
either an excited state in the right dot aligns with the ground state in the left or a hole
excited state in the left dot aligns with the ground state in the right. After tunneling
occurs the system can relax, trapping the electron in the right dot. Finally at D, no
excited states exist to match the initial configuration and allow elastic tunneling, so
there is a dip in the right-dot occupation.
Figure 6.3(a) shows the sensing signal measured throughout the pulse/dc-energy-shift
plane, with prominent diagonal steps marking the ground state transitions during
the pulse, and fine bands extending horizontally from each step reflecting excited
state transitions available during the probe time. Figure 6.3(b) presents the same
data differentiated with respect to V2 and smoothed along both axes. Here, steps
in the raw signal appear as positive ridges, and excited state peaks become bipolar.
The comparison with single-dot transport spectroscopy is clear: the pulse opens an
energy window, and as the window expands, more excited states become accessible
and emerge from the ground state feature. Features on the left (negative VE) mark
69
-20
-10
0
10
20
gls (
10-3
e2 /h)
-876 -872 -868 -864 V2 (mV) along diagonal
321
-880
-870
-860
-850
V2
(mV
) al
ong
diag
onal
-20 0 20VE pulse height (mV)
-10 0 10gls(10
-3e
2/h)
-20 0 20
-4 -2 0 2 4dg/dV2 (a.u.)
-990
-980
-970
-990
-980
-970
V2
(mV
) al
ong
diag
onal
-20 -10 0 10VE pulse height (mV)
3 2 1
(d)
(e)
V3=-1000 mV
(a) (b)
(c)
V3=-950 mV
V3=-850 mV
Figure 6.3: (a) Left sensor conductance, and (b) its smoothed derivative with respectto V2, as functions of V2 and VE. Horizontal excited state lines emerge from a diagonalground state feature as the energy window VE is increased. Vertical stripes in (a)result from sensor drift. (c) Slices of conductance (as in Fig. 6.2) as a function of V2,averaged over different VE ranges (see inset), offset vertically for clarity. Dashed curveis measured with pulses off. (d, e) Sensor conductance derivative as in (b) measuredwith the tunnel barrier either more open (d) or more closed (e) such that the probeor the reset configuration, respectively, dominates. Color scales in (d) and (e) are thesame as in (b).
70
electron excited states of the right dot or hole excited states of the left; features on
the right mark electron excited states of the left dot or hole excited states of the right.
Figure 6.3(c) shows slices of sensor conductance from Fig. 6.3(a) at three different
VE pulse heights, illustrating the expansion of the energy window while the positions
of the emerging excited states remain fixed. The dashed curve shows the transition
measured with no pulses applied. From its smooth, narrow shape, and its consistency
from transition to transition, we conclude that its width is dominated by temperature
broadening [84]. We associate this with a temperature of ∼100 mK, assuming the
electron temperature in the dot is not significantly different in the tunneling regime,
where it was previously calibrated [84]. This gives a lever arm δV2/δE = (10.5± 1)/e
relating changes in V2 (along a diagonal with V10) to changes in the energy of levels
in the left dot relative to the right. The spacing between ground-state transitions
gives the sum of the two charging energies, and assuming they are equal we find Ec =
700±70µeV. The measured excited-state gate-voltage spacing of ∼ 0.75 mV gives an
excited level spacing of ∼ 70 µeV, comparable to the ∆ ∼ 100µeV estimated from dot
area. The slightly lower measured value may reflect sensitivity to both electron and
hole excited states, giving overlapping spectra. As the energy window is increased,
excited-state-to-excited-state transitions become available, further complicating the
observed spectra. This may explain the blurring at V2 > 872 mV in curve 3.
Figure 6.3(c) shows that ground-state transitions for the pulse curves (solid) are
clearly broader than for the no-pulse curve (dashed). Broadening beyond temperature
is presumably due to both averaging traces with different VE and effects of overshoot
and settling of the pulse. However, as the probe step is long and insensitive to pulse
properties, the excited state peaks are not similarly broadened. In principle, the
71
excited state peaks are also immune to thermal broadening, their widths limited only
by intrinsic decay rates, although all peaks shown here exhibit full widths at half
maximum of at least 3.5 kT (0.3 mV), possibly due to gate noise.
Figures 6.3(d) and (e) repeat Fig. 6.3(b) with different values of gate voltage V3,
illustrating the effects of opening or closing the tunnel barrier beyond the regime of
excited-state spectroscopy. Changing V3 affects the interdot barrier both during and
after the pulse, but the system is most sensitive to the tunnel rate during the probe
time. Increasing tunneling by making V3 less negative by 100 mV (Fig. 6.3(d)) yields
single, horizontal features, as if no pulse were applied at all. This implies that τin
tprobe so the system quickly finds its ground state during the probe time regardless of
relative energy levels, making the pulse irrelevant to a time-averaged measurement.
Reducing tunneling by making V3 more negative by 50 mV (Fig. 6.3(e)) results in
diagonal features, indicating that dynamics during the pulse dominate behavior. In
the right half of this plot there is a single transition, implying that the system finds its
ground state while the pulse is on, then the barrier is closed such that τel, τin tprobe
and no further tunneling is permitted. On the left there are two diagonal features,
implying that an excited state is populated at the start of the reset pulse. This effect
is not understood at present, nor is it specific to the too-closed-barrier regime; it is
occasionally seen along with the understood horizontal excited state features.
6.4 Latched detection
We now turn to the second pulse/sense method described above, latched detection,
using two pulses on each gate as shown in Fig. 6.4(a). Figure 6.4(b) shows the
72
time -900
-890
-880
V2
(mV
) al
ong
diag
onal
-10 0 10VE probe pulse (mV)
420-2dg/dV2 (a.u.)
(a) (b)
0
0
VB
(mV
)V
E
treset
tprobe
90
140
Figure 6.4: (a) Two-pulse technique, shown schematically, including a reset and aprobe pulse, followed by a long measurement time when no tunneling is allowed. (b)Conductance derivative with respect to V2, shows excited states appearing now asdiagonal lines.
derivative of the sensor signal measured in this configuration as a function of V2
and the VE probe pulse height, using treset = tprobe = 20 ns, V resetB = 140 mV and
V probeB = 90 mV. Here we vary the probe properties rather than the reset properties as
in Fig. 6.3, so excited states appear diagonally and the reset ground state is horizontal.
Excited states measured this way are not immune to pulse properties, and as a result,
the data in Fig. 6.4 are blurred relative to Fig. 6.3. This diminished resolution is not
fundamental, and can be reduced with more accurate pulse shaping.
We thank M. J. Biercuk and A. Yacoby for useful discussions and K. Crockett for
experimental contributions. This work was supported in part by DARPA QuIST,
Harvard NSF-NSEC, and iQuest at UCSB.
73
Chapter 7
Singlet-triplet spin blockade andcharge sensing in a few-electrondouble quantum dot
A. C. Johnson, J. R. Petta, C. M. MarcusDepartment of Physics, Harvard University, Cambridge, Massachusetts 02138
M. P. Hanson, A. C. GossardDepartment of Materials, University of California, Santa Barbara, California 93106
Singlet-triplet spin blockade in a few-electron lateral double quantum dot is inves-
tigated using simultaneous transport and charge-sensing measurements. Transport
from the (1,1) to the (0,2) electron occupancy states is strongly suppressed relative
to the opposite bias [(0,2) to (1,1)]. At large bias, spin blockade ceases as the (0,2)
triplet state enters the transport window, giving a direct measure of exchange split-
ting of the (0,2) state as a function of magnetic field. A simple model for current and
steady-state charge distribution in spin-blockade conditions is developed and found
to be in excellent agreement with experiment. Three other transitions [(1,1) to (2,0),
(1,3) to (2,2), and (1,3) to (0,4)] exhibit spin blockade while other nearby transitions
and opposite bias configurations do not, consistent with simple even-odd shell filling.1
1This chapter is adapted from Ref. [113]
74
7.1 Introduction
Great progress has been made in engineering solid-state systems that exhibit quantum
effects, providing new tools for probing fundamental problems in many-body physics
as well as new device technologies. In semiconductor quantum dots, small numbers of
confined electrons can be manipulated using electrostatic gates with surprising ease
[49, 55, 114, 115]. For the case of two electrons in the dot (quantum dot ”helium”),
Pauli exclusion and exchange induce a splitting between the spin singlet and triplet
states that can be controlled by gates and magnetic fields [92, 116]. In double dots, a
consequence of this splitting is current rectification, in which transitions from the (1,1)
to the (0,2) state (ordered pairs indicate electron occupancy in each dot) is blockaded,
while the opposite bias case, involving transitions from (0,2) to (1,1) proceeds freely.
Rectification is a direct consequence of spin selection rules [117].
Spin blockade of this type can be understood by considering positive and negative
bias transport in a double dot containing one electron in the right dot, as indicated
in Figs. 7.1(b) and (c). An electron of any spin can enter the left dot, making
either a (1,1) singlet or triplet, these states being nearly degenerate for weak interdot
tunneling [118]. In contrast, the right dot can accept an electron only to make a (0,2)
singlet. At positive bias (Fig. 7.1(b)) current can flow: an electron enters the right
dot to make a (0,2) singlet, tunnels to the (1,1) singlet, and escapes. At negative
bias (Fig. 7.1(c)), an electron can enter the left dot and form a (1,1) triplet state. A
transition from the (1,1) triplet to the (0,2) singlet is forbidden by conservation of
spin and transport is blocked.
In this chapter we report a detailed investigation of spin blockade through a lateral
75
1
234567
8
9 10 11 12
GS1
ID
GS2
a)
1 µm
GS2
b) + Bias
GS2
c) – Bias
Figure 7.1: (a) Electron micrograph of a device identical in design to the one mea-sured. Gates 2–6 and 12 define the double dot, 1 and 7 form QPC charge sensors,8 separates the left QPC and double dot current paths, and 9–11 are unused. Blackspots denote ohmic contacts. The origin of spin blockade, when the left dot has 0–1electrons and the right dot 1–2 electrons, is illustrated in (b) and (c). Opposite spinsrepresent a singlet, same spins a triplet. The left dot accepts any spin but the rightcan only form a spin singlet, blocking negative bias current once the wrong spin oc-cupies the left dot. A charge sensor (GS2) also registers the blockade, as a secondelectron is sometimes in the right dot at positive bias, but not at negative bias.
few-electron double-dot system, measured using both transport and charge sensing
by a nearby quantum point contact (QPC) to detect the charge arrangement during
blockade [71, 84]. We observe that transmission through double-dot states containing
two electrons is strongly rectified, while transmission of the first electron is symmetric
in bias. Negative-bias blockade is truncated when the (0,2) triplet state enters the
bias window, allowing the magnetic field dependence of the singlet-triplet splitting to
be measured from both transport and charge sensing. Simple rate-equation models
for transport and charge distributions reproduce key features in both types of data,
and allow relative tunnel rates to be extracted. Finally, other charge transitions are
investigated, up to the second electron in the left dot and the fourth in the right dot.
The presence or absence of spin blockade is shown to be consistent with even-odd
shell filling.
76
Spin blockade of transport, arising from a variety of mechanisms, has been investi-
gated previously in quantum dot systems [115, 119, 120, 121, 122, 123, 124, 54, 125];
the mechanism responsible for the present spin blockade was investigated in vertical
structures in Ref. [117]. The lateral, gate-defined structure we investigate has some
advantages over vertical structures by allowing independent tuning of all tunnel bar-
riers so that sequential tunneling with arbitrary dot occupations can be explored.
QPC charge sensors provide additional information, including the average charge dis-
tribution when transport is absent in the spin blockade regime.
The sample, shown in Fig. 7.1(a), is fabricated on a GaAs/AlGaAs heterostructure
with two-dimensional electron gas (density 2×1011 cm−2, mobility 2×105 cm2/Vs) 100
nm below the wafer surface, patterned with Ti/Au top gates. Gates 2–6 and 12 define
a double quantum dot in which each dot can be tuned from zero to several electrons
[126]. Gates 1 and 7 define QPCs whose conductance is most sensitive to the charge
on the right and left dots respectively. Gate 8 isolates the current path of the double
dot from the left QPC, while the double dot and right QPC share a common ground.
Gates 9–11 are not energized. Current through the double dot (ID) is measured in
response to a dc voltage on the left reservoir. A small ac excitation (6 µV at 27
Hz) allows lock-in measurement of differential conductance. Conductances of QPC
charge sensors (GS1,2) are measured simultaneously with separate lock-in amplifiers
(1 nA current bias at 93 and 207 Hz). Base electron temperature is Te ∼ 135 mK,
measured from Coulomb blockade diamonds. Two devices were measured and showed
qualitatively similar behavior; data from one device are presented.
77
7.2 Spin blockade at (1,1)-(0,2)
Figure 7.2 shows ID at ±0.5 mV bias as a function of gate voltages V2 and V6,
which primarily control the energy levels in the right and left dots. Figures 7.2(a)
and (c) were measured near the conductance resonance of the first electron, with
(m,n) indicating the charge states surrounding the resonance. At positive bias, finite
current is measured within two overlapping triangles in gate voltage space, satisfying
the inequalities µR ≥ εR ≥ εL ≥ µL or µR ≥ εR + Em ≥ εL + Em ≥ µL. Here
µL,R are the chemical potentials of the leads, εL,R are the energies to add an electron
to the ground state of either dot, and the mutual charging energy Em is the extra
energy to add an electron to one dot with the other dot occupied [40]. The first
set of inequalities defines the lower, or electron triangle, where, starting at (0,0), an
electron hops through the dots from one lead to the other. The second inequalities
define the upper, or hole triangle, where, starting at (1,1), a hole hops across the dots.
Electron and hole processes involve the same three tunneling events, only their order
changes. Schematics at the top of Fig. 7.2 depict the energy level alignments at the
vertices of the electron triangle in Fig. 7.2(a). Within the triangles, current depends
primarily on the detuning ∆ = εL − εR of one-electron states, with a maximum
current at ∆ = 0, demonstrating that interdot tunneling is strongest at low energy
loss, consistent with previous studies of inelastic tunneling in double dots [127]. At
negative bias (Fig. 7.2(c)) the triangles flip and current changes sign, but otherwise
these data mimic the positive bias case (Fig. 7.2(a)).
The corresponding data with another electron added to the right dot is shown in
Figs. 7.2)(b) and (d). At positive bias, the data qualitatively resemble the one-
78
-1.105
-1.100
840
-1.11
-1.10
-1.035 -1.030
-1.035
-1.025
151050
-1.04
-1.03V
6 (V
)
-0.96 -0.95
V6
(V)
V6
(V)
V6
(V)
V2 (V) V2 (V)
a) N=1, 0.1 T+0.5 mV
c) N=1, 0.1 T-0.5 mV
b) N=2, 0.1 T+0.5 mV
d) N=2, 0.1 T-0.5 mV
(0,0)
(1,0)
(0,1)
(1,1)
(0,1)
(1,1)
(0,2)
(1,2)
SS
T
SS
T
SS
T
|ID| (pA) |ID| (pA)
Figure 7.2: (a) Magnitude of current ID as a function of V2 and V6 across the (1,0) to(0,1) transition at 0.5 mV bias. Ordered pairs (m,n) denote electrons on the left (m)and right (n) dots, with N total electrons present during interdot tunneling. The red(©), orange (5), and yellow (4) diagrams illustrate the level alignments boundinga bias triangle. The same configuration at -0.5 mV bias, (c), shows almost perfectsymmetry. (b) and (d) show the equivalent data at the (1,1) to (0,2) transition.Current flows freely at positive bias, as depicted in the green diagram (). Negativebias current is suppressed by spin blockade (blue diagram, ) except on the lower(purple diagram, ?) and upper edges. Insets to (b) and (d) show results of a rateequation model which captures most features of the data (see text).
electron case. However, at negative bias the current is nearly zero, except along the
outermost edges of the electron and hole triangles. Referring to the diagrams above
Fig. 7.2, at positive bias, current proceeds freely from right to left through singlet
states (green square). At negative bias, an electron enters the left dot into either the
79
(1,1) singlet or triplet. If it enters the (1,1) singlet, it may continue through the (0,2)
singlet. However, once an electron enters the (1,1) triplet, it can neither continue to
the right (into the (0,2) singlet) nor go back into the left lead because it is below
the Fermi level and the hole it left quickly diffuses away. Thereafter, negative-bias
transport requires a spin flip or a second-order spin exchange process with one of
the leads. Insofar as these processes are relatively slow, transport in this direction is
blockaded.
Along the outer edge of the lower (electron) triangle, where transport is observed
in the negative-bias direction (Fig. 7.2(d), purple star), an electron trapped in the
(1,1) triplet state is within the thermal window of the left lead and will occasionally
exchange with another electron possibly loading the (1,1) singlet, which can the move
to the right, through the (0,2) singlet, and contribute to current. An analogous
mechanism in the hole channel allows negative-bias current along the upper edge of
the hole triangle: with transitions from (1,1) to (1,2) within the thermal window,
the blockade created by an occupied (1,1) triplet can be lifted by adding an electron,
making a (1,2) state, then removing it, possibly leaving a (1,1) singlet that can
contribute to current.
A simple rate-equation model allows the spin-blockade picture to be quantitatively
checked against transport data, and also indicates where charge resides in the double
dot, which can be compared to charge sensing data. The model takes two degenerate
levels in the left dot, representing the (1,1) singlet and triplet states, coupled equally
to a thermally broadened left reservoir (i.e., ignoring the extra degeneracy of the
triplet2) and a single level of the right dot, representing the (0,2) singlet (assuming
2Including the full triplet degeneracy would decrease the calculated current along
80
the (0,2) triplet is energetically inaccessible) coupled to the right reservoir. The sin-
glet levels are coupled by thermally activated inelastic tunneling, with the shape of
the ∆ = 0 peak inserted to match the positive-bias current data. Temperature, mu-
tual charging energy, and the gate capacitances are determined from measurements.
Calculated current is shown in the insets to Figs. 7.2(b) and (d). The model resembles
the experimental data, with two minor exceptions: At positive bias, measured current
is higher in the hole triangle than the electron triangle, implying that the dot-lead
tunnel barriers are, in this case, more transparent with the other dot occupied. Also,
the small but finite blockade current is absent in the model, as expected since the
model contains only first-order, spin-conserving processes.
Figure 7.3 shows the charge sensor dataGS2 versus V2 and V6, acquired simultaneously
with each panel in Fig. 7.2. A plane is subtracted from each data set to remove direct
coupling between the gates and the QPC, leaving only the effect of the average dot
occupations. Away from the bias triangles we see plateaus for each stable charge state,
which are used to calibrate the response. In Figs. 7.3(a) and (c), QPC conductance
jumps ∆GR = 0.016 e2/h due to a charge in the right dot, and ∆GR = 0.008 e2/h
due to the left. These values vary between data sets, but this QPC is always about
twice as sensitive to the closer dot.
Within each bias triangle, the sensing signal varies with the fraction of time an elec-
tron spends in each charge state. Consider Fig. 7.3(a), the one-electron positive bias
the blockaded triangle edges by roughly a factor of two. While we cannot quantita-tively fit to the model given the asymmetry between the electron and hole triangles,the current amplitude with no degeneracy does seem correct to better than a factorof two. One possibility is that the singlet and ms = 0 triplet may be mixed by nuclei(see [128] and Ch. 8), making the effective degeneracies of blockaded and unblockadedstates equal.
81
-1.105
-1.100
200GS2 (10-3 e2/h)
-1.11
-1.10
-1.035 -1.030
-1.035
-1.025
100
-1.04
-1.03
-0.96 -0.95V2 (V)V2 (V)
c) N=1, 0.1 T-0.5 mV
(0,0)
(1,0)
(0,1)
(1,1)
(0,1)
(1,1)
(0,2)
(1,2)
V6
(V)
V6
(V)
V6
(V)
V6
(V)
SS
T
SS
T
SS
T
GS2 (10-3 e2/h)
b) N=2, 0.1 T+0.5 mV
d) N=2, 0.1 T-0.5 mV
a) N=1, 0.1 T+0.5 mV
Figure 7.3: Charge sensor signal GS2 measured simultaneously with each panel ofFig. 7.2. A plane is subtracted from each panel to remove direct gate-QPC coupling.The first electron, (a) and (c), again shows bias symmetry while the second, (b) and(d), is missing features at negative bias. The model used in Fig. 7.2 also reproducesthe charge sensor data, as shown in the insets to (b) and (d), with slight disagreementdue to second-order processes (see text).
data. As with transport through the dot, charge sensing is primarily dependent on
detuning, ∆. For small interdot tunneling, the system rests mainly in (0,1), thus at
large detuning the sensing signal matches the (0,1) plateau. In the electron triangle,
as detuning decreases and interdot tunneling increases, the system spends more time
in (1,0) and (0,0). Both increase the right QPC sensor conductance. In the hole
triangle, the system accesses (1,0) and (1,1), which respectively increase and decrease
82
the right sensing signal.
Assuming the same tunnel rates to the leads in the electron and hole triangles, lead
asymmetry can be quantified by comparing the rate-equation model to the sensor
signal in the two triangles. At the one-electron transition, the charge sensor signal can
be written as a deviation from the (0,1) plateau, ∆GS = ∆GR(p10 +p00)−∆GL(p10 +
p11), where pmn is the occupation probability of charge state (m,n). At positive
bias, current flows only leftward through each barrier, thus conservation of current
gives pe01ΓD = pe
10ΓL = pe00ΓR in the electron triangle and ph
01ΓD = ph10ΓR = ph
11ΓL
in the hole triangle, where ΓD, ΓL, and ΓR are the interdot, left, and right barrier
tunnel rates. As each set of three probabilities sums to unity, we can calculate each
individually (e.g. pe00 = [1 + (ΓR/ΓL) + (ΓR/ΓD)]−1). We define a quantity α as the
ratio of ∆GS in the hole triangle to ∆GS at the same detuning (thus the same ΓD) in
the electron triangle, which gives α = [(S − 1)ph10 − ph
11]/[(S − 1)pe10 + Spe
00], where S
is the QPC sensitivity ratio ∆GR/∆GL. Inserting the expressions for the occupation
probabilities in terms of tunnel rates, we find
α =S − 1− ΓR/ΓL
(S − 1)ΓR/ΓL + S. (7.1)
This expression is independent of ΓD (and therefore independent of detuning), imply-
ing that the lineshape of ∆GS as a function of detuning is the same in each triangle,
up to an overall positive or negative factor α. This can be seen qualitatively in
Figs. 7.3(a) and (c), with equally narrow ∆GS peaks of the same sign in each tri-
angle, and in Fig. 7.3(b), with equally broad ∆GS peaks of opposite sign in each
triangle. Linear cuts through these data sets confirm that the lineshapes are identical
to within measurement errors. Rearranging Eq. 7.1 gives the lead asymmetry
ΓR/ΓL =S − 1− Sα
1 + (S − 1)α. (7.2)
83
From Fig. 7.3(a), we measure α = 0.35 ± 0.05, giving ΓR/ΓL = 0.23 ± 0.08. Similar
analysis at negative bias (Fig. 7.3(c)) yields ΓR/ΓL = 0.35 ± 0.10. In the positive
bias two-electron case (Fig. 7.3(b)), the hole triangle shows a negative sensor change,
indicating that the left barrier is more opaque than the right. It is not possible to
further quantify this ratio, as we know from Fig. 7.2(b) that tunnel rates in the two
triangles differ. In this case, unambiguously determining the individual tunnel rates
would be possible only by using both charge sensors simultaneously.
In the spin blockade region (Fig. 7.3(d), near the blue diamond), no sensor variation
is seen, confirming that the system is trapped in (1,1). By including lead asymmetry
and QPC sensitivities, the rate-equation model used for the insets to Figs. 7.2(b) and
(d) yields charge sensor signals as well. These are shown in the insets to Figs. 7.3(b)
and (d), and again the agreement is good. The one discrepancy is between the two
triangles at negative bias. The model shows a thermally broadened transition from
(1,1) to (0,2) at a detuning equal to the bias. This is equivalent to strictly zero
interdot tunneling, in which case the system has a ground state and occupations
mimic a zero-bias stability diagram. However, between the bias triangles the only
mechanism for equilibration is a second order process of each dot exchanging an
electron with its lead. This is slow enough to occur on par with the second order and
spin-flip processes noted above which circumvent the blockade, so the sensor shows a
mixture of (1,1) and (0,2).
Figure 7.4 illustrates the features that arise at dc bias larger than ∆ST , the (0,2)
singlet-triplet energy splitting. Panels (a)–(d) show ID and GS2 versus voltages V2
and V6 at ±1 mV bias and a perpendicular magnetic field B⊥ = 0.9 T. At negative
bias, spin blockade is lifted when the (1,1) triplet is raised above the (0,2) triplet
84
(pink +), so in Figs. 7.4(c) and (d), current turns on and steady-state populations
shift where ∆ > ∆ST , with ∆ the detuning of the (1,1) ground state with respect to
the (0,2) ground state, given by the distance from the upper left side of the triangles.
At positive bias, current increases (Fig. 7.4(a)) and populations shift (Fig. 7.4(b))
when the Fermi level in the right lead accesses the (0,2) triplet (black ×), which
occurs at a fixed distance from the lower left side of each triangle. Thus, ∆ST can
be measured using either current or charge sensing and either sign of bias, with the
energy scale calibrated by equating the triangle size to the dc bias [40]. Current
and sensing give consistent values of ∆ST , but different splittings are measured at
different biases and gate voltages, presumably reflecting real changes in ∆ST as these
parameters are tuned. The two measurements in Fig. 7.4(a) give 480 and 660 µeV.
The negative bias measurement gives ∼ 520 ± 50 µeV in Fig. 7.4(c). The fact that
positive bias measurements differ more than negative bias measurements implies that
occupation of the left dot has a strong effect on the right dot levels.
Increasing B⊥ reduces ∆ST , as seen in previous experiments [92, 116, 117, 129, 130,
131, 111, 132, 133] and theoretical work [133, 134, 135, 136], bringing the negative
bias current threshold closer to the zero-detuning (upper left) edge. Figure 7.4(e)
shows negative bias current at B⊥ = 1.7 T, where a dramatic decrease in ∆ST is
seen compared to Fig. 7.4(c). Figure 7.4(f) shows ∆ST as a function of B⊥ based on
negative-bias data at different gate voltage settings. For the open squares, no voltages
besides V2 and V6 were changed during the field sweep. The tunnel barriers closed
with increasing field, obscuring the measurement above 2.5 T, before the splitting
reached zero. The sweep yielding the filled circles and panels (a)–(e), started from
different gate voltages and gave V4 and V12 corrections quadratic in field to keep the
85
-1.01
-1.00
-0.98 -0.97
-1.00
-0.99
300|ID| (pA)
-0.98 -0.97
0.8
0.6
0.4
0.2
0.0
∆ ST (
meV
)
210 B⊥ (T)
-1.02
-1.01
-0.99 -0.98
a) N=2, 0.9 T+1 mV
b)
c) N=2, 0.9 T-1 mV
d)
e) N=2, 1.7 T-1 mV
f)
V6
(V)
V6
(V)
V6
(V)
V2 (V)
V2 (V) V2 (V)
S
T
S
T
S
T
S
TS
T
S
TS
T
S
T
c
e∆ST
∆ST
∆ST∆ST
300GS2 (10-3 e2/h)
Figure 7.4: Measurements of the singlet-triplet splitting, ∆ST . At +1 mV bias, ID(a) and GS2 (b) as a function of V2 and V6 at the (1,1) to (0,2) transition show newfeatures where triplet current is allowed (black diagram, ×) rather than just singletcurrent (green, ). At negative bias, ID (c) turns on and GS2 (d) changes whentriplet states break the spin blockade (pink + vs. blue ). Several ∆ST values aremeasured from (a)–(d), which we attribute to gate voltage dependence of ∆ST . AsB⊥ increases, (e), ∆ST decreases rapidly. The dependence of ∆ST on B⊥ is shownin (f), with the solid and open series both taken at negative bias from different gatevoltage configurations.
86
barriers roughly constant. ∆ST dropped until at 2.1 T no splitting was observed. A
split feature was hinted at in the 2.5 T positive bias data, implying that the triplet
had become the ground state, but again the signal vanished as field increased. Zeeman
splitting, though it would be a small effect regardless, is entirely absent at negative
bias as interdot transitions connect states with equal Zeeman energy.
7.3 Spin blockade at other charge transitions
All of the results so far have concerned two charge transitions, corresponding to adding
the first electron to the left dot and either the first or second to the right. Figure 7.5
extends these measurements through the second charge transition in the left dot,
after which the tunnel barriers could no longer be balanced, and through the fourth
transition in the right dot, after which complications were observed in the honeycomb
pattern, indicative of a triple dot beginning to form. These measurements allow us
to comment on the robustness of singlet-triplet spin blockade with more particles and
more complicated energy levels.
The simplest situation we can imagine for our double dot is no symmetry to the
potential in either dot. In this case there is no orbital level degeneracy, and we
expect even-odd filling, with each odd electron inhabiting a new level and each even
electron pairing with the preceding odd electron to form a spin singlet. Figure 7.5(b)
illustrates where we would expect to see spin blockade within this model. When the
total number of electrons during interdot tunneling is odd, current flows from one
spin doublet to another. Current flows freely in either bias direction, indicated by a
dark double-headed arrow. For even total electron number, one charge state (in which
87
-0.448
-0.444
V6
(V)
(2,0)
(1,1)
c)
-0.392
-0.386
V6
(V)
(2,1)
(1,2)
d)
-0.346
-0.340
V6
(V)
(2,2)
(1,3)
e)
-0.336
-0.330
V
6 (V
)
(2,3)
(1,4)
f)
-0.448
-0.444
V6
(V)
-0.433 -0.429 V2 (V)
(2,0)
(1,1) -0.392
-0.386
V6
(V)
-0.364 -0.358 V2 (V)
(2,1)
(1,2) -0.346
-0.340
V6
(V)
-0.302 -0.296 V2 (V)
(2,2)
(1,3) -0.336
-0.330
V6
(V)
-0.270 -0.264 V2 (V)
(2,3)
(1,4)
-0.500
-0.494
V6
(V)
(1,0)
(0,1)
g)
-0.432
-0.426
V
6 (V
)
(1,1)
(0,2)
h)
-0.402
-0.396
V
6 (V
)
(1,2)
(0,3)
i)
-0.376
-0.370
V6
(V)
(1,3)
(0,4)
j)
-0.500
-0.494 V
6 (V
)
-0.486 -0.480 V2 (V)
(1,0)
(0,1) -0.432
-0.426
V6
(V)
-0.406 -0.400 V2 (V)
(1,1)
(0,2) -0.402
-0.396
V6
(V)
-0.354 -0.348 V2 (V)
(1,2)
(0,3) -0.376
-0.370
V6
(V)
-0.330 -0.324 V2 (V)
(1,3)
(0,4)
-0.55
-0.50
V6
(V)
-0.40 -0.35V2 (V)
420-2
(0,0)(0,1)
(0,2)(0,3)
(0,4)
(1,0)
(2,0)
a)dGS2/dV6 (a.u.)
ID (pA) 0 3 (c,g,h)10 (d,e,f,i,j)
(1,1) (1,2) (1,4)(1,0) (1,3)
(0,1) (0,2) (0,4)(0,0) (0,3)
(2,1) (2,2) (2,4)(2,0) (2,3)b)
Figure 7.5: Spin blockade at other charge transitions. (a) GS2 over a wide range ofV2 and V6, differentiated numerically in V6, maps out the first several charges addedto each dot. Bipolar transitions at higher occupation numbers (upper right corner)result from large dot conductance affecting the charge sensor measurement due to theshared drain reservoir. (b) Locations of current suppression due to spin blockade inan even-odd level filling model. ID vs. V2 and V6 at each of these eight transitionsfor positive (upper half) and negative (lower half) bias are shown in panels (c)–(j).Dashed white outlines are guides to the eye indicating the full bias triangle size. Oneoutline in each panel is drawn by hand, then rotated 180 and overlaid on the oppositebias data to ensure equal size. White arrows indicate the boundaries of observed spinblockade regions.
each dot contains an even number of electrons) has a singlet ground state, whereas
the other charge state (two odd occupations) may form a triplet or a singlet. Current
in one bias direction, indicated by the red X, is therefore expected to be blockaded
at interdot bias less than the energy gap to the first triplet state in the doubly even
charge state. Current at the opposite bias flows uninhibited from one singlet state to
the other.
Figure 7.5(a) shows a large scale plot of the charge sensor signal versus V2 and V6,
with zero bias across the dot, differentiated with respect to V6. Sharp features in this
88
type of plot mark changes in the occupation of each dot and allow us to determine
the absolute occupation at each point. These data were taken after applying positive
voltages to the gates while recooling the same device used in Figs. 7.2–7.4, which
shifted all charge transitions to less negative gate voltages and gave smaller singlet-
triplet splittings. Gates 3, 4, 5, and 12 are adjusted at each charge transition to
balance the interdot and dot-lead tunnel barriers. Figures 7.5(c)–(j) show current
through the dot versus V2 and V6 in a field of 100 mT for all charge transitions
involving the first or second electron in the left dot and the first through fourth in
the right dot—the same transitions shown in Fig. 7.5(b). The top half of each panel
is measured at positive bias (250 µV in (c), 500 µV elsewhere) while the bottom half
is measured at an equal negative bias. Spin blockade is seen in each configuration
predicted in Fig. 7.5(b) and nowhere else. At the (1,1) to (0,2) transition (Fig.
7.5(h)), we measure negative-bias spin blockade as in Figs. 7.2–7.4, this time with
(0,2) singlet-triplet splitting ∆ST ∼ 300 µeV. This smaller splitting, compared to the
600–900 µeV seen previously, indicates either a larger and shallower or more elongated
potential profile compared with the previous cooldown, such that the energy cost of
triplet occupation is decreased. At the nominally symmetric (1,1) to (2,0) transition
(Fig. 7.5(c)) spin blockade is observed in positive bias, and yields a (2,0) singlet-triplet
splitting which is smaller still, roughly 80 µeV.
The (1,3) to (2,2) transition, Fig. 7.5(e), again shows positive bias blockade, with a
splitting of ∼140 µeV. In this case there are two different (2,2) triplet states capable
of breaking the spin blockade. One is a triplet in the left dot and singlet in the right,
in which the conduction cycle looks identical to (1,1) to (2,0) but with a static singlet
pair added to the right dot. The second possibility is a singlet in the left dot and
89
triplet in the right. Here a singlet correlation between two of the right-dot electrons
is transferred to the left dot, just as in (1,2) to (2,1) transport, while the triplet
correlation between one electron in each dot is transferred to the two electrons in the
right dot. Because of these two available processes, the measured ∆ST is expected to
represent the minimum of the left-dot and right-dot singlet-triplet splittings.
A singlet ground state is expected in any situation where one quantum level contains a
spin pair and is not degenerate with any unfilled levels. A two-electron dot, therefore,
will always have a singlet ground state because the first quantum level, with no nodes,
never has orbital degeneracy [118]. For this reason, it is clear that the spin blockade
we observe in Figs. 7.5(c), (e), and (h) all originate from singlet ground states and
triplet excited states. In nearly symmetric potentials, however, a four-electron dot
can show a triplet ground state due to Hunds rule. This has been observed in a
vertical quantum dot through a detailed analysis of the spin transitions seen out to
high perpendicular magnetic field [49], and in a lateral quantum dot comparable in
size to one of our dots by the absence of Zeeman splitting in a large parallel field
[137].
The spin blockade observed at the (1,3) to (0,4) transition, Fig. 7.5(j), with ∆ST ∼
200–300 µeV, does not necessarily indicate that the singlet is the ground state of the
(0,4) configuration, just that singlet or triplet has an energy lower than the other
by the observed splitting. Moreover, dependence of the splitting on perpendicular
magnetic field up to a few tesla, as measured in Fig. 7.4(f), would not distinguish
between singlet and triplet ground states. If the ground state were a singlet, it would
gain more energy due to lateral confinement than the more extended triplet state, as
was the case with the two-electron dot, and therefore the singlet-triplet gap would
90
close with field. A triplet ground state would indicate near-degeneracy of two orbital
levels, so that adding a perpendicular field would favor singlet occupation of the level
with lower orbital angular momentum, and again the gap would close with field.
In the present experiment, a parallel field was unavailable and, as described above,
high perpendicular field has too strong an effect on tunnel barriers to allow measure-
ments of other spin transitions, thus a direct comparison with the methods used in
earlier works is not possible. However, one piece of evidence implies that our four-
electron dot, in contrast to the previous measurements, has a singlet ground state.
Consider the (1,4) to (2,3) transition, Fig. 7.5(f). The (2,3) configuration consists of
two singlet pairs and one free spin, making an overall spin-1/2 state. The same is
true of (1,4) if the four-electron dot is a singlet, which would not give any current
suppression, as observed. However, if the four-electron dot were a triplet, the overall
(1,4) state could have a spin of 1/2 or 3/2, leading to a suppression of current from
(1,4) to (2,3), which might be termed doublet-quadruplet spin blockade. Instead, as
much current is seen in Fig. 7.5(f) at positive bias as at negative bias, leading us to
conclude that this four-electron dot is a spin singlet. While this is the reverse of the
conclusion drawn in the two previous experiments, it is perhaps not surprising given
the triple-dot behavior seen at higher right dot occupations. This behavior indicates
an elongated potential in the right dot, possibly with a double minimum even while
it acts as a single dot.
We acknowledge useful discussions with Jacob Taylor and Amir Yacoby. This work
was supported by the ARO, the DARPA QuIST program, the NSF including the
Harvard NSEC, and iQuest at UCSB.
91
Chapter 8
Triplet-singlet spin relaxation vianuclei in a double quantum dot
A. C. Johnsona, J. R. Pettaa, J. M. Taylora, A. Yacobya,b, M. D. Lukina,C. M. Marcusa, M. P. Hansonc, A. C. Gossardc
aDepartment of Physics, Harvard University, Cambridge, MassachusettsbDepartment of Condensed Matter Physics, Weizmann Institute, Rehovot, IsraelcDepartment of Materials, University of California, Santa Barbara, California
The spin of a confined electron, when oriented in some direction, will lose mem-
ory of that orientation after some time. Physical mechanisms leading to this relax-
ation typically involve coupling the spin either to orbital motion or to nuclear spins
[138, 139, 140, 141, 142, 143, 144]. Relaxation of confined electron spin has been
previously measured only for Zeeman or exchange split spin states, where spin-orbit
effects dominate [131, 55, 11]; spin flips due to nuclei have been observed in optical
spectroscopy studies [12]. Using an isolated double quantum dot and direct time do-
main measurements, we investigate in detail spin relaxation for arbitrary splitting of
spin states. We show that electron spin flips are dominated by nuclear interactions
and are slowed by several orders of magnitude with a magnetic field of a few mT.1
1This chapter is adapted from Refs. [145, 146]
92
8.1 Introduction
The coupling of nuclear spins to electrons in low-dimensional semiconductors is known
from optical and transport studies in quantum Hall systems to yield rich physical ef-
fects and provide new probes of the relatively isolated quantum system of nuclear
spins in solids [147, 148, 149, 150]. Confined electrons interacting with relatively few
nuclei are particularly sensitive to hyperfine coupling. This can lead to dramatic
effects such as tunnelling currents that slowly oscillate in time and electrical control
and readout of nuclear polarization [151, 12]. In this chapter we show that the in-
teraction between single electrons confined in quantum dots with ensembles of lattice
nuclei can dominate electron spin relaxation.
We use high-frequency pulsed gates to measure spin relaxation in a GaAs double
quantum dot (Fig. 8.1(a)). Measurements are performed near the (1,1) to (0,2) charge
transition, where (n,m) denotes the absolute number of electrons on the left and right
dots. In the (0,2) configuration, the two electrons form a spin singlet to avoid the large
Pauli exclusion energy cost (0.4 meV kT ≈ 10 µeV) of occupying an excited orbital
state [130]. In the separated (1,1) configuration, the two electrons may occupy any
spin state. That is, apart from any Zeeman energy (∼ 2.5 µeV at 100 mT), the singlet,
(1,1)S, and three triplets, (1,1)T−, (1,1)T0, and (1,1)T+ (ms = −1, 0, 1 respectively),
are effectively degenerate, given the weak interdot coupling to which the system is
tuned.
Spin relaxation is measured by preparing an unpolarized mixture of (1,1) states and
monitoring the probability of transition to (0,2)S after the latter is made lower in
energy by changing the electrostatic gate configuration. The different local environ-
93
grs
1 µm
RL
gls
BextBnuc,l Bnuc,r
a
b
-513
-506
VL
(mV
)
-395 -390VR (mV)E
R
M
M’
M’’
-8 -4 0 grs(10-3 e2/h)
(1,1)
(0,1)
(0,2)
(1,2)
c
VR
VL
time
E R M
d
(1,1) (0,2) (0,1)
E
ST S
R
M
M'
M''(1,2)
(0,1)
e
f
Figure 8.1: Spin-selective tunnelling in a double quantum dot. a) Micrograph of adevice similar to the one measured. Metal gates deplete a two-dimensional electrongas 100 nm below the surface, with density 2×1011 cm−2. A double dot is definedbetween gates L and R. Electrons tunnel between the dots and to conducting leads.Conductances gls and grs of the left and right QPCs reflect average occupation ofeach dot. b) In (1,1), spatially separated electrons feel different effective fields fromhyperfine interaction with the local Ga and As nuclei, plus a uniform external field. c)Voltage pulses on gates L and R cycle through three configurations: empty (E), reset(R) and measure (M). d) Right sensor conductance grs as a function of dc voltageson the same two gates around the (1,1) to (0,2) transition, with pulse displacementsshown by points E, R, and M . Dashed lines outline the (0,1), (1,1), (0,2), and(1,2) charge state plateaus during step M . Inside the solid-outlined ‘pulse triangle,’the ground state is (0,2), but higher sensor conductance indicates partially blockedtunnelling. A plane is subtracted from the raw data to remove direct gate-QPCcoupling. e) Energetics of the pulse sequence. In (0,2), only the singlet is accessible,whereas in (1,1), singlet and triplet are degenerate. (0,1) and (1,2) are spin-1/2doublets. Step E empties the second electron, then R loads a new electron into theleft dot, occupying all four (1,1) states equally. At M , (0,2)S is the ground state, butonly (1,1)S and the ms = 0 triplet (1,1)T0 can tunnel. Mixing of (1,1)T+ and (1,1)T−with the singlet is weak away from zero field, so their tunnelling is blocked. f) At M ′,(0,1) has lower energy than (1,1) and provides an alternate, spin-independent pathto (0,2). At M ′′, (1,2) provides this alternate path.
94
ments acting on the two spins cause the two-electron spin state to evolve in time,
and only if this spin state passes near (1,1)S is a transition to (0,2)S allowed. The
average occupancy of the left dot, which reflects the probability of this transition, is
monitored using the QPC charge sensors. Conductances gls and grs of the left and
right sensors change by several percent when an electron enters the dot nearest the
sensor [71, 84, 91, 126].
8.2 Pulsed-gating technique
The energy levels of each dot were controlled by voltage pulses on gates L and R, as
shown in Fig. 8.1(c). The double dot was cycled through three configurations, de-
picted in Fig. 8.1(e), while measuring the average QPC conductances. In the ‘empty’
(E) step, the second electron is removed, leaving a (0,1) state. In the ‘reset’ (R) step,
a new second electron is added, initializing the (1,1) state to an unbiased mixture of
the singlet, (1,1)S, and the three triplets (1,1)T−, (1,1)T0, and (1,1)T+. In the ‘mea-
surement’ (M) step, (0,2) is lowered relative to (1,1) until (0,2)S becomes the ground
state, while the (0,2) triplets remain inaccessible, above the (1,1) states. Because
tunnelling preserves spin, only (1,1)S can relax to (0,2)S, while the (1,1) triplets are
spin-blockaded from making this transition [117, 113].
The measurement step accounted for 80% of the pulse period (E and R were each
10%) so the time-averaged charge-sensor signal mainly reflects the charge state during
the measurement time, tM . Figure 8.1(d) shows grs as a function of the d.c. offsets to
gate voltages VL and VR with pulses applied. The dashed lines indicate locations of
ground-state transitions during the M step, as seen in unperturbed double dots [84].
95
a) b)
∆V
VR=25 mVP
-415 -400VR (mV)
-520
-530
VL
(mV
)
-415 -400VR (mV)
-5 50dgrs/dVL (a.u.)
Figure 8.2: Gate pulse calibration. a) dgrs/dVL as a function of VR and VL aroundthe (1,1) to (0,2) charge transition. Charge states are labeled (M,N), where M (N)is the number of electrons on the left (right) dot. b) 25 mV pulses with a 50% dutycycle and 10 µs period are applied to the coax connected to gate R. This results intwo copies of the charge stability diagram shifted relative to one another by ∆~V .
Gate pulses alter this signal only within the ‘pulse triangle’ (outlined by solid white
lines). Here grs is intermediate between the (0,2) and (1,1) plateaus, indicating that
although (0,2) is the ground state, the system is often stuck in the excited (1,1) state.
In the regions labelled M ′ and M ′′, alternate, spin-independent relaxation pathways,
shown in Fig. 8.1(f), circumvent the spin blockade.
Gates L and R are connected via bias tees to dc voltage sources and to pulse gen-
erators through coax cables with ∼20 dB of inline attenuation.2 Pulse heights are
calibrated by applying pulses to a single gate and measuring the charge stability di-
agram. Figure 8.2(a) shows the charge sensor signal vs. d.c. gate voltages VL and VR
with no pulses applied, differentiated with respect to VL. Because of the differenti-
ation, steps in the sensor signal (marking boundaries between ground state charge
configurations, as seen e.g. in Fig. 8.1(d)) appear as dark or bright lines. Figure 8.2(d)
2We use Anritsu K251 bias tees. Tektronix AWG520 arbitrary waveform genera-tors (with ∼ 1 ns rise times and 10-bit vertical resolution) are used for high speedmanipulation of the gate voltages.
96
shows the same charge stability diagram acquired with square pulses applied to gate
R (25 mV pulse height before the attenuators, 50% duty cycle, period τ=10 µs).
This results in two copies of the charge stability diagram, the right-most (left-most)
charge stability diagram reflects the ground state charge configuration during the low
(high) stage of the pulse sequence. The gate-voltage offset between them, ∆~V , is
used to calibrate pulse amplitudes. For this pulse, ∆~V is not purely horizontal as one
might expect (since it is applied solely to gate R, the horizontal axis), but is angled
upward slightly, indicating that there is more cross-coupling between gates at a.c.
than at d.c. Additional calibrations are performed for gate L, which primarily shifts
the honeycomb in the vertical direction (not shown). A linear combination of pulses
on gates R and L can then be used to shift the stability diagram in any direction in
gate space.
8.3 Spin relaxation measurements
Our spin relaxation measurement technique relies on the fact that (1,1)T to (0,2)S
transitions are spin blocked. Two control experiments demonstrate that the observed
time dependence of the charge sensing signal is due to spin blocked transitions rather
than simply slow inelastic interdot tunneling. The first is to compare the pulse se-
quence described above (Fig. 8.1) with a reversed pulse sequence, where the M point
is in (1,1) while the R point is in (0,2). Now tunneling from R to M should always
proceed on a time scale set by the interdot tunnel coupling, since the (0,2)S to (1,1)S
transition is not spin blocked. Although these data are not shown, no signal is seen
in the pulse triangle for this reversed sequence unless tM is dramatically reduced,
97
a) B=1.2 T b) B=1.4 T
c) B=1.6 T d) B=1.8 T-512
-516
-393-396
VL
(mV
)
VR (mV)
0 5 10 grs (10-3 e2/h)
-393-396 VR (mV)
-512
-516
VL
(mV
)
J
JJ
Figure 8.3: grs as a function of VR and VL for increasing B (which reduces J), withτM=10 µs. Eventually (0,2)T is lowered into the pulse triangle [inset of (b)]. At thispoint, (1,1)T to (0,2)T transitions are energetically possible and the transition fromthe (1,1) to (0,2) charge state is no longer “spin blocked”. This cuts off the tip of thepulse triangle in (0,2), see (b). A best-fit plane has been subtracted from the data in(a)–(d).
confirming that long persistence time of the forward pulse sequence is not simply due
to a long inelastic tunneling time.
The second control experiment measures signal in the pulse triangle as a function of
a large magnetic field B. Throughout this experiment B is applied perpendicular to
the sample plane, and as shown in Ch. 7, we expect that such a field will decrease
the (0,2) singlet-triplet splitting from ∼ 800 µeV at B = 0 to zero when B ∼ 2 T.
Figure 8.3(a) shows grs as a function of VL and VR while applying the forward pulse
sequence with B = 1.2 T and τ = 10 µs. For these data, the M -point detuning where
(0,2)T has lower energy than the (1,1) states resides outside of the pulse triangle
98
(J > EM , the mutual charging energy) and the (1,1)T to (0,2) transitions are spin
blocked. For B = 1.4 T (Fig. 8.3(b)) the (0,2)T state is low enough in energy that
the (1,1)T states can directly tunnel to the (0,2)T manifold at high detunings. Now
(1,1) to (0,2) tunneling can proceed, and there is no longer a (1,1) charge signal in
the (0,2) region of the pulse triangle at high detuning. This cuts off the tip of the
pulse triangle. The spin-blocked region continues to shrink as B is increased. From
these data, we find J ∼340, 280, and 180 µeV for B=1.4, 1.6, and 1.8 T respectively,3
consistent with the values derived from d.c. transport in Fig. 7.4(f).
For smaller fields, the B and tM dependence of the charge sensor signal is shown in
Fig. 8.4. With tM = 8 µs, a large signal is seen in the pulse triangle, indicating that
some of the (1,1) to (0,2) transitions are spin blocked. As tM is increased this signal
decreases (Fig. 8.4(b)), indicating that tM is approaching the (1,1) singlet-triplet
relaxation time. This is accompanied by a reduction in the pulse triangle size due to
thermally activated processes as in Fig. 8.1(f). Similar data, but at B = 0, are plotted
in Figs. 8.4(c) and (d). The signal in the pulse triangle is noticeably weaker for the
same tM , particularly near the (1,1)-to-(0,2) charge transition, indicating enhanced
spin relaxation.
Detailed measurements of residual (1,1) occupation as a function of detuning (the
energy difference between the (1,1) and (0,2) states) are shown in Fig. 8.5. Conduc-
tances gls and grs were measured along the diagonal white line in the upper panel of
Fig. 8.5, for various values of B and tM , and converted to occupation 〈N〉 by scaling
to the average (1,1) and (0,2) levels outside the pulse triangle. Data are shown in
3The gate-voltage to energy conversion is determined from finite-bias triangles asdescribed in Ch. 7.
99
-394 -391
d B=0
-511
-507
-394 -391VR (mV)
c B=0
b B=100 mT
-511
-507
VL
(mV
)
-8 -4 0 grs (10-3 e2/h)
a B=100 mTtM=8 µs
(1,1)
(0,1) (0,2)
(1,2)
tM=8 µs
tM=80 µs
tM=80 µs
VL
(mV
)
VR (mV)
Figure 8.4: Dependence of the occupancy of the (1,1) state on measurement time,tM , and external field, B. a) Charge sensor conductance, grs, as a function of VL andVR with short pulses (tM = 8 µs) at B = 100 mT. Large average occupation of (1,1)is seen throughout the pulse triangle. Near the triangle edges, thermally activatedtunnelling to the leads allows fast relaxation to (0,2), (see Fig. 8.1(f)). b) For longerpulses (tM = 80 µs), thermally relaxed triangle edges expand towards the centre of thetriangle. c) At B = 0, the (1,1) occupation is extinguished at low detuning (near the(1,1)-(0,2) charge transition) as tunnelling to (0,2) becomes possible from the (1,1)T+
and (1,1)T− states. d) Combine these two effects at zero field with long pulses, andno residual (1,1) occupation is seen, indicating complete relaxation to (0,2).
detail for the points labelled A through D. As in Fig. 8.4, strong field dependence was
found at low detuning (point A), where inelastic interdot tunnelling dominates. This
field dependence vanishes at higher detuning where thermally activated tunnelling to
the leads dominates.
As in previous work [141, 12], we model spin evolution in (1,1) by treating the ensem-
100
1.0
0.5
0.0
<N
>
tM (µs)
1.0
0.5
0.0
<N
>
1.0
0.5
0.0
<N
>
1.0
0.5
0.0
<N
>
-50 0 50B (mT)
10 100
0 mT150 mT
8 µs80 µs800 µs
A
B
C
D
-460
-457
VL
(mV
)
-369 -365VR (mV)
AB
CD
Γin~104 ΓT
Γin~100 ΓT
Γin~5 ΓT
Γin< ΓT
(1,1) (0,2) (0,1)(1,2)
Figure 8.5: Detailed measurements of blockaded (1,1) occupation. Average occupa-tion 〈N〉 of the (1,1) charge state, based on calibrated charge sensor conductances,at four detuning points (labelled A, B, C, D in the uppermost panel). Left panelsshow 〈N〉 as a function of tM at B = 0 and B = 150 mT. Middle panels show 〈N〉as a function of B for different tM times. Diagrams at right show schematically therelative position of energy levels and the extracted ratios of inelastic (Γin) to thermal(ΓT ) decay rates.
101
ble of nuclear spins within each dot as a static effective field Bnuc with slow internal
dynamics, that adds to any applied Zeeman field (see Fig. 8.1(b)). Bnuc is randomly
oriented with r.m.s. strength Bnuc = b0√I0(I0 + 1)/Nnuc, where b0 = 3.5 T is the
hyperfine constant in GaAs, I0 = 3/2 is the nuclear spin and Nnuc is the effective
number of nuclei with which the electron interacts [139, 140, 152, 153]. In our dots,
Nnuc is estimated at 106–107, giving Bnuc ≈ 2–6 mT. The spins precess in a character-
istic time tnuc = ~/g∗µBBnuc ≈ 3–10 ns, which can be regarded as an inhomogeneous
dephasing time T ∗2 . At B = 0, all four (1,1) spin states mix in this time, and tun-
nelling will appear insensitive to spin. With B > Bnuc, however, only (1,1)T0 and
(1,1)S are degenerate. These will continue to mix with the same rate, but (1,1)T+
and T− will be frozen out.
8.4 Analysis and discussion
To model this mixing, we assume static nuclear fields during each pulse, a spin-
preserving inelastic interdot tunnelling rate Γin from (1,1)S to (0,2)S, and a spin-
independent rate ΓT due to thermally activated tunnelling via the (0,1) and (1,2)
charge states (see Ap. A for details). Zeeman eigenstates for two spins in fields
Bz + Bnuc,l and Bz + Bnuc,r, denoted |(1, 1)s, s′ (s, s′ = ±1/2), decay to (0,2)S on
the basis of their overlap with (1,1)S, with rates Γs,s′ = Γin|〈(1, 1)S|(1, 1)s, s′〉|2 as
long as Γin g∗µBBnuc. Averaging over nuclear field configuration and short-time
dynamics gives decay rates for the T±-like states:
Γ± 12,± 1
2=
Γin
4(1 + (B/Bnuc)2)(8.1)
102
and Γ± 12,∓ 1
2= Γin/2 − Γ± 1
2,± 1
2for the S-like and T0-like states. At B = 0, total
transition rates for all (1,1) states into (0,2)S are the same, τ−10 = Γin/4 + ΓT . For
B > Bnuc, transition rates τ−1B = Γ± 1
2,± 1
2+ ΓT from (1,1)T± to (0,2)S are suppressed
by field, while transitions from (1,1)S and (1,1)T0 to (0,2)S are accelerated by up
to a factor of two because they no longer mix with (1,1)T±. During the gate-pulse
transition from R to M , the relatively fast transition from (1,1)S to (0,2)S allows a
fraction q of the (1,1)S state to transfer adiabatically to (0,2)S, reducing the initial
occupation of (1,1)S. The resulting average occupancy N of (1,1) after a time tM is:
N(tM) =1
tM
∫ tM
0
dt(1
2e−t/τB +
2− q
4e−t(2τ−1
0 −τ−1B )) (8.2)
Experimentally measured values for N as functions of tM and B for various detun-
ings are shown in Fig. 8.5, along with fits to the above theory. An additional field-
independent parameter, N∞, accounts for non-zero N(tM) at long times owing to
thermal occupation of (1,1). N∞ is zero at large detuning but increases, as expected,
near zero detuning. Non-zero q values are found only at very low and very high de-
tuning (where the R point is near zero detuning), where the slew rate of the pulse is
low as it crosses to (0,2). With these parameters and τ0 set for a given detuning by
fitting the zero-field data (gray), the high-field data (black) are fitted with just the
longer decay times τB for the (1,1)T± states. The field-dependence curves are then
fully determined by Bnuc, which is most accurately determined from data in Fig. 8.6,
as discussed below. Drift in sensor conductance over long field sweeps is compensated
by allowing a vertical shift in the field-dependence curves. The depth and width of
the dips in these curves are not adjustable.
Figure 8.6 shows the extracted decay times τ0 and τB versus detuning for various
103
fields. As the magnetic field increases, more points at high detuning fall along a
line in this semi-log plot, denoting exponential energy dependence as expected for
a thermally activated process. This persists over three orders of magnitude at the
highest field, and with calibration from transport measurements yields a temperature
of 160±20 mK. At zero field, thermal decay dominates only at the highest detunings,
and the low-detuning times are well fitted by a power-law function of detuning with
exponent 1.2±0.2 and offset 700 ns, typical of inelastic tunnelling in double quantum
dots [127]. Adding these two processes gives the zero-field theory curve in Fig. 8.6, in
good agreement with the zero-field data. The 10-mT curve is fitted using these zero-
field parameters, but with times for the inelastic component increased by the factor
(1 + (B/Bnuc)2) from Eq. 8.1. The fit gives Bnuc = 2.8± 0.2 mT, or Nnuc ≈ 6× 106,
within expectations. This value uniquely determines the remaining theory curves. For
τB longer than about 1 ms the decay is faster than theory predicts (though still 103
times slower than at B = 0), indicating that another mechanism such as spin-orbit
coupling may operate on millisecond timescales [131, 55, 11]. Spin-orbit coupling is
expected to dominate spin relaxation at external fields of several tesla [55]. This
regime is better suited to parallel fields, which couple almost exclusively to spin, than
to the perpendicular orientation used here, which affects orbital wavefunctions at high
fields.
Given Bnuc above, the model predicts an inhomogeneous dephasing time T ∗2 ≈ 9 ns
for this device, which is independent of external field despite the enhanced relaxation
times measured at higher fields. Up to 1 ms, the excellent agreement between ex-
periment and theory indicates that hyperfine interaction is the only relevant source
of spin relaxation in this system. Several strategies are available to circumvent this
104
deca
y tim
e (s
)
0.2 0.6 1.0 1.4∆VR (mV) along diagonal
10-6
10-5
10-4
10-3
10-2
A B C D
150 mT
70 mT
30 mT
10 mT
0 mT
Figure 8.6: Decay of (1,1) occupancy as a function of detuning at various magneticfields. Dotted lines mark the locations of points A through D from Fig. 8.5. Fit ofzero field theory (lowest curve) to data sets all fit parameters except Bnuc, which isdetermined by fitting to the 10-mT data. Theory curves at other fields are then fullydetermined. Error bars at zero field result from the least-squares fitting. Error barsat non-zero field reflect changes in the resulting decay rate as the zero-field fittingparameters are varied within their uncertainties.
short dephasing time. Materials with zero nuclear spin, such as carbon nanotubes,
avoid hyperfine effects entirely. Controlling Bnuc via nuclear polarization [12, 151] is
tempting, but high polarization is required for T ∗2 to increase substantially [154]. An
alternative is to use spin echo techniques such as pulsed electron spin resonance to
extend coherence to the nuclear spin correlation time, expected to be of the order of
100 µs in these devices [141].
We thank H. A. Engel and P. Zoller for discussions. This work was supported by the
ARO, the DARPA-QuIST programme, and the NSF, including the Harvard NSEC.
105
Chapter 9
Singlet separation and dephasing ina few-electron double quantum dot
J. R. Pettaa, A. C. Johnsona, J. M. Taylora, A. Yacobya,b, M. D. Lukina,C. M. Marcusa, M. P. Hansonc, A. C. Gossardc
aDepartment of Physics, Harvard University, Cambridge, MassachusettsbDepartment of Condensed Matter Physics, Weizmann Institute of Science, Rehovot,
IsraelcDepartment of Materials, University of California, Santa Barbara, California
Quantum computation and information processing requires generation and manipula-
tion of entangled states. While several recent experiments show that spin relaxation
times (T1) in GaAs quantum dots can approach several ms, a direct time-resolved
measurement of the spin dephasing time T ∗2 , has to date not been reported. Here we
measure T ∗2 for single electrons confined in a gate-defined double quantum dot using
time-domain interferometry of correlated electron pairs. The measured T ∗2 = 10 ns is
well accounted for by a model of hyperfine interactions with the GaAs nuclei.1
1This chapter has not appeared in print. An expanded version including morecomplex Rabi and spin-echo pulse sequence measurements is in press at Science.
106
9.1 Introduction
Coupled GaAs quantum dots allow controlled isolation of one or more electrons
[115, 25]. These devices may be well suited for quantum information processing
because the tunnel couplings, exchange interactions, and electron occupation can be
easily changed by tuning gate voltages and external magnetic fields [101]. In addition,
the time it takes for a single spin to relax (T1) has been shown to be very long, making
electron spin a good candidate for quantum information storage [131, 55, 11, 146].
Coherent manipulation of spin states for quantum computation requires that the spin
dephasing time T ∗2 be much longer than the typical gate operation time. A measure-
ment of this figure of merit has to date not been reported for spins in quantum dots.
Knowledge of this timescale is crucial because a short T ∗2 will likely hinder some appli-
cations of coherent spintronics [155]. Time-resolved Faraday rotation measurements
in bulk GaAs systems have demonstrated spin dephasing on 100 ns timescales, but
this timescale cannot be directly applied to quantum dots because the mechanisms for
spin dephasing have different strengths in confined systems [156, 144, 47]. Pioneering
optical spectroscopy measurements and charge sensing experiments have measured
the relaxation of electron spin polarization at zero field (See [12] and Ch. 8). A direct
time-resolved measurement of T ∗2 for a single spin in a quantum dot is lacking.
Two potentially important sources of spin dephasing are spin-orbit and hyperfine
(electron-nuclear spin) interactions. Theoretical estimates indicate that spin-orbit
coupling should not cause rapid spin dephasing of electrons in GaAs dots, giving,
for this mechanism, T2 = 2T1 ∼ 1 ms [144]. On the other hand, dephasing due to
the hyperfine interaction may be quite rapid [139, 157, 140, 141, 142]. In a simpli-
107
fied picture, the hyperfine interaction in a GaAs quantum dot creates an effective
magnetic field, Bnuc = b0√I0(I0 + 1)/Nnuc, where b0 ∼ 3.5 T is the hyperfine con-
stant in GaAs, I0 = 3/2 is the nuclear spin of all species, and Nnuc is the effective
number of nuclei with which the electron interacts [152]. As discussed below (and in
Ch. 8), Nnuc ∼ 6 × 106 in our devices, giving Bnuc ∼ 3 mT. Since these fields are
uncorrelated in different quantum dots, two spatially separated electron spins will
dephase on a timescale tnuc =√
3~/(g∗µBBnuc) ∼ 10 ns, which can be regarded as an
inhomogeneous dephasing time T ∗2 [157].
We directly measure the spin dephasing time in an isolated GaAs double quantum
dot, fabricated on a GaAs/AlGaAs hetrostructure grown by MBE, using pulsed gates
and quantum point contact (QPC) charge sensing [71, 91, 126], as shown in Fig. 9.1.
QPC conductance, gs, measured as a function of the gate voltages VL and VR, is
plotted in Fig. 9.1(b) and maps out the double dot charge stability diagram. Labeling
charge states with the integer pair (n,m) (the second index indicating the sensed right
dot), we focus on transitions involving the (0,2) and (1,1) states, where previous
experiments have demonstrated spin-selective tunneling (see [146, 117] and Ch. 7).
The measurement protocol is illustrated in Fig. 9.1(c). We first initialize the system in
the (0,2) singlet ground-state (denoted (0,2)S), which is separated by ∼400 µeV from
the first triplet state of (0,2). At time t = 0, the double-well potential is tilted to make
the (1,1) charge state the new ground state. This induces a transition from (0,2)S into
the spatially separated (1,1) singlet state, (1,1)S. For B = 0 and interdot tunneling,
tc, tuned to be weak,2 such that (1,1)S is nearly degenerate with the three (1,1) triplet
2Tunnel coupling is chosen stronger here than in Ch. 8 so that charge may beadiabatically transferred between dots (described in detail below), in contrast to thepurely inelastic transfer in Ch. 8.
108
x
V(x
)
1 µm
0=t
=t τS
A C
RL
B evlovE
erusaeM
S
)2,0()1,1(
ST
eraperP
)2,0()1,1(
S
ST
etarapeS
)2,0()1,1(
S
ST
)2,0()1,1(
S
ST
gs
D
824-
234-
V R )Vm( 004-404-
gs
(10
-3e
2/h
)0
01
)1,0()2,0(
)1,1()2,1(
VL
(mV
)
Figure 9.1: a) Scanning electron micrograph of a sample identical to the one used inthis experiment. Voltages on gates L and R control the number of electrons in theleft and right dots. The QPC conductance, gs, is sensitive to the number of electronsin the double dot. b) gs measured as a function of VL and VR reflects the doubledot charge stability diagram. Charge states are labeled (m,n), where m(n) is thenumber of electrons in the left(right) dot. c) Measurement protocol. The double dotis initialized in the spin singlet, (0,2)S. At t = 0, the spin singlet is spatially separatedinto (1,1)S. The separated electrons are left to evolve in the presence of hyperfine andspin-orbit interactions until t = τs. At t = τs we make a projective measurement bytilting the double well potential so that (0,2)S becomes the ground state. (1,1)T to(0,2)S transitions are spin blocked, while (1,1)S to (0,2)S transitions proceed freely.The QPC detects spin-blocked transitions. d) Energy level configuration at each stageof the experiment.
states, (1,1)T+, (1,1)T0, and (1,1)T− (representing ms = +1, 0,−1), mixing between
all spin states occurs rapidly. Splitting off the (1,1)T+ and (1,1)T− states with an
external Zeeman field allows rapid mixing only between (1,1)S and (1,1)T0. At time
t = τs, the double-well potential is tilted to again make (0,2)S the ground state. This
projects the evolved (1,1) spin state onto (0,2)S. Due to spin selection rules, only
(1,1)S can tunnel to (0,2)S, while (1,1)T to (0,2)S transitions are blocked. Failure
109
to return to the (0,2) charge state, as reflected in the QPC conductance, indicates
that the return transition was spin blocked, and hence dephasing occurred during the
time τs [146]. This procedure is analogous to a time-domain interferometer, where
two correlated electrons are separated, recombined, and measured.
To implement this protocol, voltage pulses were applied to gates L and R using a
two-channel arbitrary waveform generator. The four-step pulse sequence is shown
in Fig. 9.2(a). The system is initialized into (0,2)S by waiting at point P for 200
ns. Then (0,2)S is transferred to (1,1)S by pulsing to point S, passing through an
intermediate point P ′ for 200 ns to reduce pulse overshoot. The gates are then held at
point S for a duration τs followed by a spin-state projection measurement by pulsing
to point M , where (0,2)S becomes the ground state, and holding for 9.4 µs before
the cycle is repeated. Spin-blocked transitions are detected by measuring the time-
averaged QPC conductance gs, which reflects the configuration at point M , since
> 95% of the cycle time is spent there.
9.2 The spin funnel
Interdot tunnel coupling, tc, causes a splitting of (0,2)S and (1,1)S, lowering the
energy of the hybridized (1,1)S–(0,2)S state by t2c/(2|εs|) to leading order in tc/εs (i.e.
far from degeneracy), where εs is the energy shift away from the (0,2)–(1,1) charge
degeneracy point. Because of this splitting, the (1,1)T+ state will cross the hybridized
(1,1)S at finite Zeeman field, such that εs = t2c/(2g∗µBB), as illustrated in Fig. 9.2(c).
In the vicinity of this degeneracy, rapid mixing of (1,1)S and (1,1)T+ states can occur,
and the probability of getting stuck in (1,1) after the projective measurement onto
110
B
ε sVL
V R
P
M, P’
S
VL S
M, P’
P
0
B (mT )
gµB
1(,1
)T -
1(,1
)T +1(,1
)T 0
(1,1)S
(0,2)S
ygr
en
E
εS0
B (mT)-100 1000
-2
-3
0
-1
0
(0,1)
(0,1)
(1,1)
(1,1)
(0,2)
(0,2)
(1,2)
(1,2)
A C E
B D
0 2 4 gs (10 -3 e2/h)
F
ε s)
Vm(
ε s)
Vm(
∆MS
“spin funnel”
(0,2)S
|S> |T +>
|S> |T 0>
Figure 9.2: Spin dephasing in a double dot. a) Pulse cycle P → P ′ → S → M → Pfor a small separation ∆MS between the M and S points. b) Pulse sequence for alarge ∆MS. c) Finite interdot tunnel coupling results in a tunnel splitting at zerodetuning. Tuning εs probes different regions of this energy level diagram. For small∆MS (resulting in a small εs), (1,1)S and (1,1)T+ are in resonance at finite field.A larger ∆MS (more negative εs) probes dephasing between (1,1)S and (1,1)T0 athigh field. At B = 0, (1,1)S can mix with all three (1,1)T states, enhancing theprobability to remain in a spin blocked configuration. d) Schematic grey-scale plotof the probability to remain in a (1,1)T spin blocked state, P(1,1)(τs T ∗2 ) derivedfrom the energy level diagram in (c). The spin funnel corresponds to mixing between(1,1)S and (1,1)T+ and results in P(1,1)(τs T ∗2 ) = 1/2. For very negative εs, (1,1)Sto (1,1)T0 dephasing occurs at finite field, resulting in a broad band of signal withP(1,1)(τs T ∗2 ) = 1/2. At B = 0, (1,1)S can mix with all three triplet states andP(1,1)(τs T ∗2 ) = 2/3. e) gs measured as a function of εs and B using the pulsesequence illustrated in (a), ∆MS = (−1, 1.05) mV, τs = 200 ns. f) gs measured as afunction of εs and B using the pulse sequence illustrated in (b), ∆MS = (−3, 3.15)mV, τs = 200 ns.
(0,2)S is P(1,1)(τs T ∗2 ) = 1/2. For more negative εs, such that |εs| > t2c/(2g∗µBBnuc),
different behavior is expected in different field regimes. With B > Bnuc, (1,1)S and
(1,1)T0 will approach degeneracy, resulting in P(1,1)(τs T ∗2 ) = 1/2. For B < Bnuc,
all four (1,1) spin states approach degeneracy. In this regime P(1,1)(τs T ∗2 ) = 2/3.
These regimes are illustrated Fig. 9.2(d) in a schematic gray-scale plot representing
P(1,1)(τs T ∗2 ) as a function of εs and B. The resonance condition set by (1,1)S
111
0
0.0 0.2 0.4 0.6 P (1,1)
-1
-2
ε S)
Vm(
B (mT)-100 100-50 500
Figure 9.3: Composite plot of P(1,1)(τs T ∗2 ) measured as a function of εs and Bwith τs = 200 ns made by combining three separate field sweeps, each with a different∆MS. For −1.5 < εs < 0 mV, a well-defined spin funnel is seen. With εs < −1.5mV, the zero-field feature is flanked by a signal with roughly half the amplitude thatextends out to the highest measured fields. Overlay: a power law fit to the spin funnelgives ε ∝ −|B|−0.4.
and (1,1)T+ results in the characteristic 1/B dependence of the spin funnel. Mixing
between (1,1)S and (1,1)T0 states results in a field-independent band of P(1,1) signal
at large negative εs, where |εs| > t2c/(2g∗µBBnuc). Within this band, P(1,1) is further
enhanced at low fields, B < Bnuc, reflecting mixing of (1,1)S with all three (1,1)T
states.
A finite singlet-triplet splitting in (0,2) limits the εs measurement range for a given
pulse sequence. To span a wide range of εs we take several data sets with different
112
separations ∆MS between the S and M points. Figure 9.2(e) shows a plot of gs as a
function of B and εs with ∆MS = (−1, 1.05) mV in (VL, VR) and τs = 200 ns. For small
|εs| we observe the spin funnel, as described above. For large |εs|, (probed by setting
∆MS = (−3, 3.15) mV) we find a field-independent signal, augmented at B = 0, as
shown in Fig. 9.2(f). A composite image of P(1,1) created by combining three data
sets with different ∆MS and τs = 200 ns is plotted in Fig. 9.3. The observed spin-
blocked signal is in good qualitative agreement with the expected structure illustrated
in Fig. 9.2(d). Fitting the spin funnel data to a power law ε ∝ −|B|α gives a power
α = −0.4 rather than the expected α = −1, behavior which is not understood
at present but which is confirmed by more recent Rabi oscillation and spin-echo
experiments.
9.3 Measurement and theory of T ∗2
The spin dephasing time is determined by measuring the spin blocked signal as a
function of τs. We set ∆MS = (−7, 7.35) mV, so that −7 < εs < −6 mV. P(1,1)(τs)
is plotted in Fig. 9.4 for both B = 100 mT, where the (1,1)S–(1,1)T0 dephasing
process is measured, and B = 0 mT, where mixing between (1,1)S and all three
triplet states is explored. The average of ten such measurements is shown to reduce
signal noise. We find that P(1,1) increases on a ∼10 ns timescale and saturates for
τs > 20 ns at values of ∼0.3 for B = 100 mT and ∼0.5 for B = 0 mT. These results
are qualitatively consistent with expectations: For τs T ∗2 , dephasing is minimal,
leading to P(1,1) ∼ 0, i.e., (1,1)S does not have time to evolve into a state that will
get stuck in (1,1). As τs reaches T ∗2 , P(1,1) should rise, since the (1,1)S state will mix
113
τS )sn( 05040302010
P (1
,1)
0.0
2.0
4.0
6.0
T 2 sn 01=*
tnemirepxETm 0=B
Tm 001=ByroehT
Tm 0=B
Tm 001=B
Figure 9.4: P(1,1) measured as a function of τs. At B = 100 mT, only (1,1)S to (1,1)T0
dephasing is allowed and P(1,1) approaches ∼0.3 for long τs. For B = 0 mT, (1,1)Scan mix with all three (1,1)T states and approaches ∼0.5 for long τs. Theory predictslong-τs values of 0.5, 0.67. The ratios of the experimental P(1,1) values are consistentwith this expectation to within < 20%, but the overall amplitudes are reduced (seetext). The data are fit using a model that accounts for distinct nuclear environmentsin the left and right dots. Best fits to the data give Bnuc = 2.3 mT and T ∗2 = 9.6 ns.
significantly with degenerate (1,1)T states.
To model the experimental P(1,1)(τs) curves, we calculate the probability for (1,1)S to
dephase in the presence of two distinct quasi-static nuclear fields acting on the spins
in the left and right dots. The resulting functional forms are:
P(1,1)(t) =
C1
(1− 1
2(1 + e
− t2
t2nuc )
), B Bnuc
C2
(1− 1
3(1 + e
− 12
t2
t2nuc (1− t2
t2nuc) + e
− t2
t2nuc
(1− t2
t2nuc
)2), B = 0
(9.1)
Fitting parameters are C1, C2, and tnuc. Fitting to the 100 mT data, we find tnuc =
9.6± 0.5 ns and C1 = 0.62± 0.01. For the B=0 data we set tnuc = 9.6 ns and find a
best fit C2 = 0.74 ± 0.01. Theory predicts an overshoot in P(1,1) for B=0, which is
114
an overdamped remnant of the Rabi oscillations of the electron spin in the hyperfine
field. The experimental data do not exhibit this overshoot. More complicated theories
may be required to account for this discrepancy. Theory also predicts C1 = C2 = 1
for 100% population transfer efficiency and perfect initialization. The reduced C1
and C2 values found in experiment may result from a residual exchange energy j or
non-adiabatic population transfer. In addition, we expect C1 C2. Experimentally,
these values differ by < 20%. This discrepancy may be due to switching noise in the
100 mT data (evident in the data in Fig. 9.3), which upon signal averaging pulls down
the P(1,1) signal level, or finite exchange j. Nevertheless, the field dependence of P(1,1)
suggests that only the (1,1)T0 state is being accessed at 100 mT whereas all three
triplet states are accessed at B = 0. Since the time dependence predicted by Eq. 9.1 is
not a simple exponential form we define T ∗2 = tnuc, with a best fit value of T ∗2 = 9.6 ns,
corresponding to Bnuc = 2.3 mT. In principle, T ∗2 could be enhanced by polarizing the
nuclear spins or taking advantage of their long correlation time. Significant increases
in T ∗2 would require nearly full nuclear polarization, beyond current capabilities [140,
147]. The anticipated long nuclear correlation time compared to both gating times
and electron spin relaxation and dephasing times suggests that approaches based on
fast estimation of nuclear fields or echo techniques should be effective.
We acknowledge useful discussions with Hans-Andreas Engel, Emmanuel Rashba, and
Peter Zoller. Funding was provided through the ARO under DAAD55-98-1-0270 and
DAAD19-02-1-0070, the DARPA-QuIST program, and the NSF under DMR-0072777
and the Harvard NSEC.
115
Appendix A
Hyperfine-driven spin relaxation
This Appendix details the interaction between nuclei and electrons in a two-electron
double quantum dot, as used to derive the theoretical curves for Ch. 8. This section
is primarily the work of Jacob Taylor, and appeared as supplemental online material
for Ref. [145]
A.1 Definition of Bnuc
We start by reviewing the effective magnetic field picture for nuclear spins in a single
quantum dot [141], then extend it to the double-dot case. For a single electron in
a single quantum dot with large orbital level spacing, i.e., ~ω |g∗µBB|, we can
write a spin Hamiltonian for the ground orbital state of the dot, ψ(r) and neglect
higher orbital states. Including the Zeeman interaction, HB = −g∗µBB · S, and the
hyperfine contact interaction, HHF = Av0
∑k |ψ(rk)|2Ik · S, we can define an effective
(Overhauser) magnetic field due to nuclei,
Bnuc =Av0
−g∗µB
∑k
|ψ(rk)|2Ik , (A.1)
116
such that Htot = HB + HHF = −g∗µB(B + Bnuc) · S completely determines the
spin hamiltonian for a single electron quantum dot (b0 = A−g∗µB
= 3.47 Tesla for
GaAs [152]). At temperatures T ~γn|B|kB
, where γn is the gyromagnetic ratio for a
given nuclear spin species, the equilibrium state of each nuclear spin is well described
an identity density matrix, ρk = 1/(2I0 + 1); correspondingly, the equilibrium ex-
pectation values for Bnuc are, for large N , equivalent to a Gaussian variable with no
mean and a root-mean-square width,
Bnuc =√〈|Bnuc|2〉 = b0[v
20
∑kk′
|ψ(rk)|2|ψ(rk′)|2〈Ik · Ik′〉]1/2
= b0
√v0I0(I0 + 1)
∫d3r |ψ(r)|4 . (A.2)
Furthermore, as the internal dynamics of the nuclear field, determined by the nuclear
Zeeman energy and Knight shift, are substantially slower than g∗µBBnuc, to a good
approximation the field is quasi-static, i.e., does not change during the electron-
nuclear spin interaction time [141].
For a lateral quantum dot [158] with a Fock-Darwin ground-state wavefunction in xy
of rms width σ =√
2~m∗ω
and a quantum well confinement wavefunction in z of width
l,
ψ(r) = [16
l3/2z e−4z/l][
exp(− (x2+y2)4σ2 )
√2πσ2
] (A.3)
and Bnuc = b0√I0(I0 + 1)
√3v0
8πlσ2 = 2.32 Tesla√
v0
lσ2 = 18 mT√
ω[meV]l[nm]
. By compar-
ison to the homogeneous case for a dot of volume V (ψ(r) = V −1/2), we can also
define an effective number of nuclear spins, N = [v0
∫d3r |ψ(r)|4]−1 = 8πlσ2
3v0, such
that√〈|Bnuc|2〉 =
b0√
I0(I0+1)√
N.
For a double quantum dot with one electron in each well, in an external field with
117
negligible exchange coupling (J = 0), the spin hamiltonian is
H(1,1) = −g∗µB[B(1) · S(1) + B(2) · S(2)] , (A.4)
where B(i) = Bext + B(i)nuc is the effective magnetic field in dot i. The eigenstates
are spins aligned and anti-aligned with these two fields, |s, s′〉, with eigenenergies
Ess′ = −g∗µB(B(1)s + B(2)s′) and s, s′ = ±12. For clarity, we denote eigenstates of
only the external magnetic field, e.g., with Bnuc = 0, as∣∣s, s′⟩.
A.2 Hyperfine-driven decay
We take the quasi-static approximation, where we assume that over the course of
each cycle of the experiment (one pulse sequence) the nuclear field is static [159].
This requires that the correlation time of the nuclear field be greater than 100 µs,
consistent with order kHz linewidths for solid-state NMR on GaAs [152]. We remark
that in the existing literature describing relaxation due to nuclei in a single quantum
dot [139, 140, 142, 143], the relaxation mechanism we analyze below (spin mixing due
to nuclei followed by a direct inelastic decay) is absent, requiring instead a higher-
order virtual process for spin relaxation.
We consider an inelastic tunneling mechanism that couples only to charge, leaving the
spin state unchanged. In the absence of spin selection rules, the inelastic tunneling
takes the charge state (1,1) to (0,2) with a rate Γin, which is a function of device
parameters and energy difference between the two charge states (ESG); such inelastic
decay, due, for example, to phonons, has been studied in detail elsewhere [127]. When
~Γin |Ess′|, we may adiabatically eliminate nuclear spin degrees of freedom. As the
118
ground state of (0,2), |G〉, is a spin singlet, the decay rate of a state |s, s′〉 in the (1,1)
charge configuration is Γss′ = Γin| 〈s, s′|S〉 |2, i.e., the probability overlap of the state
|s, s′〉 with the (1,1) spin singlet state, |S〉, which may then decay via charge-based
inelastic decay to |G〉.
We now evaluate the overlap matrix elements, | 〈s, s′|S〉 |2. A state |s〉 for a magnetic
field B = (x, y, z) (and of norm n = |B|) can be expressed in the original basis (|s〉,
spin aligned with the z-axis) as
|s〉 =−(2s)i(n+ z) |s = s〉 − (ix− 2sy) |s = −s〉√
2n2 + 2nz. (A.5)
In the tilde-basis, |S〉 = (∣∣∣ 12,− 1
2
⟩−
∣∣∣− 12, 1
2
⟩)/√
2. The overlap elements are then∣∣∣∣⟨S| ± 1
2,±1
2
⟩∣∣∣∣2 =[(n1 + z1)x2 + x1(n2 + z2)]
2 + [(n1 + z1)y2 + y1(n2 + z2)]2
8(n21 + n1z1)(n2
2 + n2z2),(A.6)∣∣∣∣⟨S| ± 1
2,∓1
2
⟩∣∣∣∣2 =[(n1 + z1)(n2 + z2) + x1x2 + y1y2]
2 + [−x1y2 + y1x2]2
8(n21 + n1z1)(n2
2 + n2z2). (A.7)
The states with the same |ms| value (in the instantaneous, s, basis) have the same
overlap. Averaging over all quasi-static field values, we can find 〈Γss′〉, the effective
decay rate for an experiment with many different realizations, each with a different
field value drawn from the field distribution. If the external field is aligned along
one axis, i.e., z-axis, then 〈zi〉 = B, 〈xi〉 = 〈yi〉 = 0, 〈(µi − δµzB)(νj − δνzB)〉 =
δijδµνB2nuc/3. By taking advantage of the factorization of expectation values for dots
1 and 2, e.g., 〈f(B1)g(B2)〉 = 〈f(B1)〉〈g(B2)〉, we find
〈Γ± 12,± 1
2〉 = Γ〈F 〉〈G〉 (A.8)
〈Γ± 12,∓ 1
2〉 =
Γ
2(〈F 〉2 + 〈G〉2) (A.9)
where
〈F 〉 =
⟨(n+ z)2
2n2 + 2nz
⟩=
⟨n+ z
2n
⟩119
〈G〉 =
⟨x2 + y2
2n2 + 2nz
⟩=
⟨n2 − z2
2n(n+ z)
⟩=
⟨n− z
2n
⟩are averages over a single dot. We have assumed both dots to be of equal size, i.e.,
both nuclear fields have the same effective strength, but are independent, though this
has little quantitative effect on the behavior of the system for differences on the order
of 10% or less, and no qualitative effect for differences up to 100%. The integral, over
a gaussian corresponding to the single-dot nuclear field distribution, is
I = 〈 zn〉 =
1
[2πB2nuc/3]3/2
∫ ∞
−∞dz
∫ ∞
0
r dr
∫ 2π
0
dθz exp[− r2+(z−B)2
2B2nuc/3
]√r2 + z2
. (A.10)
After integration over θ, r and variable change, u =√
3/2 zBnuc
,
I = e−3B2/2B2nuc
∫ ∞
−∞du u e
√6uB/BnucErfc(|u|) . (A.11)
We note that I ' 〈z〉√〈n2〉
= |B|√B2
nuc+B2to better than 2% for all external field values,
B. Using this approximation, we find the effective decay rates, used in the main text,
from the four eigenstates of the (1,1) charge configuration to the (0,2) singlet state to
have two forms, one for the |ms| = 1 states (T+,T−), with a rate Γ± 12,± 1
2= Γin
4B2
nuc
B2nuc+B2 ,
and the other for the |ms| = 0 states (T0, S), with a rate Γ± 12,∓ 1
2= Γin
2[1− B2
nuc
2(B2nuc+B2)
].
A.3 Thermal component
In addition to the inelastic decay from the (1,1) singlet to the (0,2) singlet, coupling to
the leads allows for a spin-independent transition of either (1, 1) → (1, 2) → (0, 2) or
(1, 1) → (0, 1) → (0, 2) to occur, breaking blockade and reducing the expected signal.
As a thermally activated process, the rate for each should dependend on the energy
difference between the (1,1) state and either (1,2) or (0,1). Denoting this detuning
120
ET , the corresponding decay has the expected form, ΓT = Γ0e−ET /kBT . We note that
larger (1,1) to (0,2) detuning ESG corresponds to smaller ET detuning, as the M point
moves closer to the top of the triangle of Fig. 8.4. This functional form is consistent
with the observed high detuning behavior shown in Fig. 8.6. Combining this decay
with the previous section results, it is convenient to define τ−10 = Γin/4 + ΓT , the
zero-field decay time, and τ−1B = Γ± 1
2,± 1
2+ ΓT , the decay rate of the |ms| = 1 states.
121
Appendix B
Electronics and wiring
B.1 DC and RF wiring in a dilution fridge
Sample wiring for a dilution fridge requires special techniques and careful attention
to balance the often contradictory demands of high electrical conductivity and low
thermal conductivity. These often conflict explicitly, because as temperature de-
creases, the conduction electrons tend to become the dominant carriers of heat. The
approach we have used to cool DC signals1 is long resistive wires, heatsunk through
their insulation at many different temperatures by wrapping the wires tightly around
a copper spool bolted to a cooling stage of the fridge.2 We used 3 meter wire looms
from Oxford, with 18 constantan wires (∼200 Ω each). The looms we bought had 6
copper wires as well which we did not use, and it was important to cut these between
heatsinking stages to avoid transmitting extra heat down the fridge. Below the mix-
ing chamber, the heatsunk resistor idea was implemented more systematically and
1In practice, DC includes frequencies up to several kHz used by lockin amplifiers.2I secured the wires to the spool with dental floss only. When Oxford does the
wiring they also use GE varnish, but this doesn’t seem necessary in practice andmakes the wiring much harder to reconfigure.
122
integrated with radiation shielding as described in Ap. C.
Some experiments called for high frequency (up to ∼1 GHz) electrical signals applied
directly to gates on the sample. The technique I used for cooling high frequency
coax cables without losing bandwidth is just one of many tried first in our group by
Leo DiCarlo. Attenuators are inserted at various temperature stages to gradually
thermalize the inner conductor, then at the bottom, just before the signal gets to the
chip, a DC voltage is added via a bias tee. A bias tee is essentially just a capacitor in
the RF line and an inductor in the DC line, such that only signals below the frequency
cutoff (∼10 kHz, although this drops when the signal is applied to a high-impedance
load like a gate3) are passsed from the DC line and only signals above the cutoff are
passed from the RF line. Then, because the RF line is already AC coupled, a DC
block 4 is inserted at the top of the fridge to explicitly minimize the power dumped
into the attenuators. Figure B.1 shows schematically how the DC and RF signals get
to the sample.
The attenuator/bias tee technique is suitable only for certain applications. You can-
not put very large RF voltages on the sample without the energy dumped in the
attenuators exceeding the cooling power of the fridge. You also cannot use this tech-
nique to measure a high-frequency response from the sample, because the response is
attenuated on the way up just as it is going down. Within these limitations, however,
3We found that although the bias tee frequency cutoff moved way down whenapplied to a high-impedance load, the sample had a strange impulse response whichpartially decreased an applied signal after several hundred µs, which we could onlyattribute to sloshing of electrons in the 2DEG. For this reason, it is recommendedthat future samples have as little of the gate over 2DEG as possible, especially forspin experiments where long pulses are crucial.
4A DC block is also essentially just a capacitor. Be careful which brand is used toget a low frequency cutoff. The best one we found was from Picosecond Pulse Labs,with a cutoff below 1 kHz.
123
RoomTemp
1KPot
700 mKStill
25 mKMix. Ch.
Sample
SS/BeCu85 mil
Nb/Nb85 mil
Nb/Nb85 mil Cu/Cu
34 mil
copperwire
20 dB
Bias Tee
AC
DCSUMDC block
Signalsource 10 dB 6 dB
10 kΩ
4K
Breakoutbox
Copperspools
Coldresistors
ColdFinger
Figure B.1: DC and RF wiring in the Nahum fridge shown schematically. One RFline and 4 DC lines are shown (3 pure DC and one as the DC component of the RFline), although the actual setup had two RF lines and 32 DC lines in two looms (seetext). See Ap. C for details of the cold finger design.
this technique provides reasonably cold electrons (∼100 mK, twice their temperature
with no coax) using robust off-the-shelf components.
Careful consideration should be given to the materials used in the coax. Coax cables
follow more or less a square-root-frequency attenuation curve, so even small cable
attenuation can change pulse shapes, but this must be balanced with the higher
thermal conductivity which normally comes with low attenuation. Above 1K there is
a long distance to the port at room temperature, and the 1K pot can handle a lot of
heat, so the material of choice for the coax is beryllium-copper (BeCu), with much
lower thermal conductivity than pure copper but nearly as good frequency response.
Due to limited availability, however, we used coax with stainless steel (SS) outer,
BeCu inner. The lines are anchored twice below 1K (at the still and mixing chamber)
before passing through the bias tee and ending up at the sample. Below the bias tee
we want both thermal conductivity and frequency response to be maximal, so regular
Cu/Cu coax is used. There is some question what material is best for the two short
segments (1K–still and still–mixing chamber). We used niobium superconducting
coax for both on the theory that the superconducting state excludes heat transfer,
124
100K
100Ω
10Ω909K
2.2mF
DC
AC
DC AC103 +
105
100K
100Ω
10M
DC
AC
DC AC103 +
105
b)a)
Figure B.2: a) AC+DC box schematic. This circuit divides the DC component by103 and the AC component by 105, and adds the below-1-Hz portion of the DC inputto the above-1-Hz portion of the AC input. b) Proposed purely resistive adder circuitwith no frequency response.
although Leo recently calculated that SS is actually better above 3He temperatures.
The choice of attenuation values is important as well. At 1K, again because the 1K
pot is a powerful cooler but also because the temperature difference (300K to 1K) is
greatest there (the outer conductor is thermalized at 4K as well, but only attenuation
affects the inner), a large attenuation is used at this point. At the still a smaller
attenuation is used, but not as small as the temperature difference (1K to 700 mK)
would indicate, since the still has much more cooling power than the mixing chamber.
Even less attenuation is used at the mixing chamber to balance heating the mixing
chamber with heating the sample directly.
B.2 AC+DC box
To measure differential conductance in the presence of a DC bias, we need a DC
voltage up to a few mV with an AC oscillation of a few µV added to it. The Agilent
33250 function generator can do this by itself, but it is cheaper and often more
convenient to use a DAC channel to generate the DC voltage, and add it with a passive
125
circuit to the oscillating voltage generated by the lockin amplifier, while dividing them
both down several orders of magnitude. The circuit I used for this (schematic shown
in Fig. B.2(a)) is a slight modification of the one designed by Duncan Stewart [50]
but with a lower cutoff frequency so that it needn’t be “tuned” to a particular lockin
frequency. There has been some discussion (prompted by Amir Yacoby via Jeff Miller)
that an even simpler purely resistive circuit (shown in Fig. B.2(b)) can do the same
addition with no frequency response and no big capacitor to worry about getting
the right kind. The only reason I can think of that one might not want to use this
simpler circuit is that the capacitor may be performing useful noise filtering functions:
any high-frequency noise in the DC bias is suppressed, including quick steps due to
changing the bias, and any DC offset in the oscillating signal is suppressed as well.
B.3 Grounding
Grounding and noise reduction are somewhat of an art. In every experiment I’ve
done there have been different things necessary to minimize noise, and sometimes an
absolutely crucial step one time becomes absolutely forbidden the next. Nevertheless,
there is one principle that usually leads pretty close to the correct grounding setup: no
ground loops. Obviously most every piece of electronics wants some part grounded,
and if something that should be grounded is left floating the measurement can go
horribly wrong. In addition, it’s not always easy to tell if a particular component
is grounded. Sometimes there are solid connections between the wall ground, the
equipment rack, the chassis, and the shields of front panel connectors, but in many
cases these connections are broken, intentionally or not.
126
Cables: signal data power
IthacoIin Vout
AWG520
GPIBout1out2
DMMGPIBVin
BreakoutBox
SMAPort
Fridge
Monitor
DACBreakout
Box
BatteryBox
Lockin
Vosc
AinBin Vout
GPIBOpticalIsolator
Out
In
Computer
NI6703DAC
GPIB Card
IsolationTransformer
2-32-3
DMMGPIBVin
DMMGPIBVin
MagnetSupply
GPIBout
Lockin
Vosc
AinBin Vout50/50
Gnd Iso
SumAC
DC
RBias
50/50
ShieldCut
Magnet
Current-bias Measurement
Voltage-biasMeasurement
Figure B.3: Full grounding configuration for an experiment, showing all signal anddata (GPIB) cables and some power cables. The thick dotted line separates quietelectronics near the fridge from potentially noisy electronics near the computer. Ide-ally this separation would be a shielded room wall, although in the case of the Nahumfridge it is open space by the doorway, with a cable trace running over the door. Typ-ical current- and voltage-bias measurement setups are shown. The lockin oscillatorand Ithaco output shields are partially grounded, so 50/50 (50Ω in both signal andground lines) or Gnd Iso boxes (ground isolation - pass the inner conductor but com-pletely break the shield) are used on these terminals. The AC+DC box is describedin Sec. B.2. The computer is grounded only through the ground wires in the DACcable, which then passes through the DAC breakout box to the fridge. The shieldof this cable is cut, the GPIB signals are run through an optical isolator, and thecomputer, monitor, and GPIB isolator are powered through an isolation transformer,with 2-3 prong converters inserted to break ground loops within this isolated system.All other electronics are plugged directly into the wall (cabling not shown).
Figure B.3 illustrates the grounding configuration I used for a complete experiment,
including the National Instruments 6703 DAC (digital to analog converter) board
which resides inside the computer. The Marcus lab has nearly standardized on a
DAC built by Jim McArthur in Harvard’s electronics shop and designed (at least
in its first incarnation) by Jim and Andrew Houck and Jeff Miller. This DAC has
127
excellent noise and grounding properties and is external to the computer so that (at
least in its later incarnations) it robustly holds its voltages regardless of what the
computer is doing (i.e. crashing). This last benefit notwithstanding, it is possible to
use an in-computer DAC with clean results and no ground loops. The trickiest part of
this setup, and the hardest problem to diagnose, was that the computer chassis and
the DAC board had slightly different ground potentials, with only a small resistance
between them, and the difference was a function of how hard the computer’s main
processor was working. The solution was to completely float the computer, and cut
the shield on the DAC cable, which was connected to the computer chassis ground
rather than the ground wires in the cable. Another particularly frustrating ground
loop may occur due to the magnet power supply, if the magnet leads make electrical
contact with the fridge structure, leading to a sample bias which varies with magnet
current. Besides careful isolation of the magnet leads inside the fridge, it is important
not to let too much water or ice collect at the top of the leads, as this can conduct.
B.4 Divider/adder for NI6703 DAC
The National Instruments 6703 is a 16-bit DAC, but we discovered in testing that
with appropriate filtering (∼100 ms time constant) its noise is twenty times smaller
than one bit This means that by chaining two channels together we can obtain an
effective 20-bit DAC (although the linearity is not good enough to cover the entire
range this way—you would use this feature by setting the coarse channel in the middle
of the desired range, then sweeping only the fine channel while taking data). This
extra resolution didn’t end up being necessary for any of the measurements I did, but
128
Vin,i
Vin,i+8
10K 10K
200K
Vout,i
Vout,i+8
Vin,i
-(Vin,i + 20
)Vin,i+8
Figure B.4: Divider/adder circuit for use with NI6703 DAC card. This is one ofeight identical elements (where i runs from 0–7) on the board in the breakout boxfor the NI6703 board. Inputs Vin,i and Vin,i+8 come from the SCSI-type connector onthe DAC cable, and the outputs Vout,i and Vout,i+8 are BNC connectors on the frontpanel. The toggle switch between straight-through and linked output modes is alsolocated on the front panel.
I built a box to allow you to, by flipping a switch, change between two independent
channels and one composite channel. As the NI6703 has 16 output channels, the first
8 are the coarse channels, the latter 8 are the fine channels, and each pair (chans. 0
and 8, chans 1 and 9, etc.) has a switch to allow linked or independent operation.
Jim’s electronics shop has all of the software and tools to design and fill a printed
circuit board, and Jim is so helpful that this, the first PCB I ever made, worked
exactly right and I even felt like I did most of it myself. Figure B.4 shows the circuit
diagram for one pair of channels, which was then repeated eight times on the board.
The board is mounted inside the DAC breakout box shown in Fig. B.3.
129
Appendix C
Cold finger design
I joined the Marcus group at a time when the number of cryostats in the lab jumped
from two to five: one dilution fridge and one 3He fridge from Stanford were joined by
a new dil fridge, a new 3He fridge, and a third dil fridge inherited from the retiring
Prof. Nahum. Add to that an old design which would break sample wires with every
few thermal cycles, and a push in the lab toward ever more complicated devices with
demand for more sample wires (which meant switching from a 28-pin to a 32-pin chip
carrier and socket), and it was clear that there was a lot of cold electronics work to
be done. Ron and I were tasked with creating a new cold finger design, with the
understanding that we would each use one (I was revamping the Nahum dil fridge
and Ron the old Stanford dil fridge) and another would replace the rickety cold finger
supplied by Oxford in the new 3He fridge. The mechanical design was mostly Ron’s
work, partly mine, and based on an idea (cold resistors, see below) of Charlie’s.
The cold finger serves several purposes. First, it holds the sample in the center of
the magnet. Some experiments call for the large field from the solenoid magnet to
be perpendicular to the sample plane to affect orbital wavefunctions, while other
130
experiments require a field parallel the sample plane so that field couples almost
exclusively to spin. Because of this, we designed two interchangeable mounts for the
socket, to be swapped without any wiring or changes to the socket itself. The parallel
field mount needs fine control over the sample angle, so that the perpendicular field
remaining from an 8–10 Tesla magnet is within the capability of the small (∼250 mT)
homemade split coil perpendicular magnet [160] to null out. Experiments in a large
perpendicular field are insensitive to small angle changes, so the perpendicular field
mount can be much simpler.
The second purpose of the cold finger is to thermally link the sample to the fridge’s
mixing chamber while shielding it from warmer radiation, and to provide the last
stage of cooling for the sample wires. At low temperatures, the wiring is the only
thing capable of carrying much heat away from the sample, so cooling the sample
and cooling the wires are essentially the same goal. Innumerable techniques have
been tried by our group and others over the years for cooling wires without making
electrical contact to them. The solution that was in place when I joined the group
(and which garnered quite a few entries in the “Don’t look in here Charlie” drawer
by Ron and Josh trying to reproduce and improve it) was the metal powder filter,
in which thin, resistive wires are sent through a bath of epoxy loaded with enough
copper or other metal powder that it almost conducts (but not quite). The problem
with this, besides the fact that it’s awfully messy and difficult to work with very
thin wires, is the tendency of differential thermal contraction to break these wires
occasionally, so that after a few years of regular thermal cycling half of the wires
didn’t work (or more frustratingly, would work at room temperature but not cold).
What we tried instead, and worked surprisingly well despite its simplicity and robust-
131
Resistor plateswith notches topass coax
Bottomcan
Perp. fieldmount
Par. fieldmount
Screws into mixingchamber or 3He pot
Cover
Pivot pin
2 screws for fineangle adjustment(2nd hidden)
Chip carrier socket(wiring pins not shown)
Notch to bring inwires and coax
6 in
Figure C.1: Schematic and photograph of cold finger. The main structure of the coldfinger is copper for optimal thermal conductivity, with a brass cover (removable halfcylinder) to minimize eddy current heating. Wires pass through resistors embeddedin thick brass plates while the coax (thermalized elsewhere) thread through notches inthe plates (offset to block line-of-sight radiation) to get to the sample. The bottom canis brass and shields the sample from warmer radiation outside. Photograph (courtesyof J. B. Miller) shows the cold finger in the 3He system, with coax installed and athread adapter for a larger bottom can.
132
ness, was to simply add regular resistors in the lines, anchored tightly into a brass
plate which seals the sample space below it from the radiation environment above.
In as far as heat is high-frequency noise in the wire, the resistors along with their
stray capacitance RC-filter this noise out. In the dilution fridges we used three of
these brass plates in series, in the 3He fridge just one. The resistor values we chose
to be 100 Ω per plate in half of the lines (to be used for ohmics) and 5 kΩ per plate
in the other half (to be used for gates). Below the plates, the radiation environment
is sealed by a screw-on brass can. With this setup in the Nahum fridge, we measured
electron temperature ∼45 mK at a base temperature of ∼25 mK.
Fig. C.1 shows some schematics of the design and a photograph of cold finger in use
on the 3He system with the cover and can removed. Since making the original cold
finger, a larger bottom can has also been made to accomodate extra coax cables to
the sample, and the adapter needed to thread on the larger bottom can is shown in
the photograph.
133
Appendix D
Igor routines
For experiments, the Marcus lab runs almost exclusively Igor Pro, a superb data
acquisition, analysis, and graphics platform1, with its own built-in programming lan-
guage. Perhaps the only concrete result of the first few experiments I did (Aharonov-
Bohm rings and Coulomb drag, neither of which yielded anything I deemed worth
putting in this thesis) is a set of higher-level Igor routines to make data acquisition
easier and to automate some common analysis functions.
The idea was to make a generic data acquisition routine: there are lots of parameters
to change in an experiment, but at any given time you will be sweeping one or two
of them, measuring one or more quantities, and plotting the results. When I started,
there was a different routine for each combination of parameters to sweep and the
data to be measured was hard coded. Making simple changes required changing many
routines, and doing even a slightly more complicated type of data acquisition would
take completely new code. I’m not sure if I really reduced my own coding time (It
took me quite a bit of time to develop this code!) but I do think these routines made
1Igor is produced by WaveMetrics (http://www.wavemetrics.com/)
134
it much easier to try new ways of taking data, which became invaluable as we started
pulsing (and varying all of the pulse heights and times) and charge sensing (measuring
up to four signals simultaneously).
The routines revolve around two generic acquisition routines, do1d for single-para-
meter sweeps, and do2d for two-parameter sweeps (although over the years several
variations have been added for specialized applications, such as averaged or interlaced
two-parameter sweeps to reduce noise). These functions take a string input for each
sweep parameter which is then fed to a routine, setval, which contains all of the
code to interface to the instruments to set physical parameters (gate voltages, field,
pulse waveforms, etc.). In addition, global variables set by a graphical interface panel
describe what is being measured by which DMMs (digital multimeters), and this
information is fed to the routine getdata, which polls the dmms and calculates the
data values to store.
A second advantage of using string inputs for the swept parameters and data types is
that it allows the program to automatically name the new data waves2 in such a way
that the user can read off what was measured and what was swept. Analysis routines
can also parse the name for this information (via the routine getlabel), allowing
them to automate tasks like labeling axes and color scales. Finally, each wave name
gets a serial number appended to distinguish it from other waves of the same type.
If several data streams are measured simultaneously, they all get the same serial
number, as do any derivative waves (slices, averages, differentiations, etc.) generated
by analysis routines. A complete wave name might look something like condbc5 212,
2“Wave” is Igor’s name for an array of data (1–4 dimensions) with a bit of extrascaling and other information tacked on.
135
which means conductance measured with field as the outer loop and DAC channel 5
as the inner loop, with a serial number of 212.
One final important aspect of my routines is automatic logging. At the beginning of
every sweep, do1d and do2d save all of the system parameters that they have access
to into a text file for later reference (although if I were doing this again today I might
save this info in the wavenote for each data wave). In the lab book then, I would write
just the serial number and notes about what appeared in the data, but not what was
measured or swept or what the other system parameters were during the sweep—a
significant time savings and incredibly useful weeks or months later when you want
to repeat a particular measurement or know everything about it for analysis.
My routines evolved over time into five files with lots of cross references such that you
needed all five to do anything. This got unnecessarily confusing and cumbersome,
so as I was writing this I condensed the routines into two files (and cleaned up the
code a bit while I was at it). alexanalysis.ipf has wave management, analysis,
and image processing routines, and it can stand on its own for data analysis uses.
alexdata.ipf, has the core data acquisition routines, and must be used along with
alexanalysis.ipf as well as with the driver files for any instruments you intend to
use.3 Copies of these alexanalysis.ipf and alexdata.ipf should be available on
the Marcus group website (marcuslab.harvard.edu) under “member information.”
To illustrate the ideas above, on the following pages I’ve included do1d (but not the
nearly identical do2d), getdata, setval, and getlabel. Finally, I’ve included two
3In particular, in order to use the gotolocal feature in alexdata.ipf, you needmy version of DMM procedures, alexDMM procedures.ipf. This sends the DMMsto local mode (continuously displaying measurements) when not in a sweep, but keepsthem in remote mode (so they go faster and don’t flash a lot) during sweeps.
136
functions to fix an annoying aspect of exporting EPS files from Igor, from the file
alexepsfix.ipf.
//do1d: generalized 1D sweep - figures out from idstr what to sweepfunction do1d(idstr,start,stop,numdivs,delay)
string idstr // see setval for allowed idstr’svariable start // starting valuevariable stop // ending valuevariable numdivs // number of points minus 1variable delay // seconds of delay between points
variable numpts=numdivs+1,nw=nextwave()string/G S_datatype1,S_datatype2,S_datatype3,S_datatype4string wave1,wave2,wave3,wave4variable d1=0,d2=0,d3=0,d4=0 // flags (0 or 1) for if data streams are in usevariable/G V_rampbackflag // flag(0 or 1) for whether to ramp back to
// initial values at the end of the sweep
// make, name, scale, and display the data wavesif(!stringmatch(S_datatype1,"-"))
sprintf wave1,"%s%s_%d",S_datatype1,idstr,nwmake/o/n=(numpts) $wave1=NaN;wave w1=$wave1setscale/i x start,stop,"",w1showwaves(wave1);movewindow/I 0,0,3,2.5d1=1
else // need something in wave1 for logwavesprintf wave1,"???%s_%d",idstr,nw
endifif(!stringmatch(S_datatype2,"-"))
sprintf wave2,"%s%s_%d",S_datatype2,idstr,nwmake/o/n=(numpts) $wave2=NaN;wave w2=$wave2setscale/i x start,stop,"",w2showwaves(wave2);movewindow/I 3,0,6,2.5d2=1
endifif(!stringmatch(S_datatype3,"-"))
sprintf wave3,"%s%s_%d",S_datatype3,idstr,nwmake/o/n=(numpts) $wave3=NaN;wave w3=$wave3setscale/i x start,stop,"",w3showwaves(wave3);movewindow/I 0,3,3,5.5d3=1
endifif(!stringmatch(S_datatype4,"-"))
sprintf wave4,"%s%s_%d",S_datatype4,idstr,nwmake/o/n=(numpts) $wave4=NaN;wave w4=$wave4setscale/i x start,stop,"",w4showwaves(wave4);movewindow/I 3,3,6,5.5d4=1
endif
logwave(wave1) //save conditions of sweep to a text file
137
nolocal() //prevent DMMs from returning to local mode after each reading//function is in "alexDMM Procedures.ipf"
variable t1=ticks,t2,secs,i=0
// data taking loopmake/o/n=(numpts) gate;gate[]=start+p*(stop-start)/numdivssetval(idstr,gate[0]);wait(3*delay)variable/G V_gtrip=0 // reset the abort sequence for I-bias measurementsdo
setval(idstr,gate[i]);wait(delay)if(d1)
w1[i]=getdata(1) //get new data into the appropriate wavesendifif(d2)
w2[i]=getdata(2)endifif(d3)
w3[i]=getdata(3)endifif(d4)
w4[i]=getdata(4)endifi+=1;doupdate// fix the scaling if time is the axisif(stringmatch(idstr,"time")||stringmatch(idstr,"noise"))
t2=ticks;secs=(t2-t1)/(60.15*i)if(d1)
setscale/p x,0,secs,w1endifif(d2)
setscale/p x,0,secs,w2endifif(d3)
setscale/p x,0,secs,w3endifif(d4)
setscale/p x,0,secs,w4endif
endifwhile((i<numpts)&&(V_gtrip==0))if(V_rampbackflag==1) //check whether to ramp back to start
setval(idstr,gate[0])endifendsweep(t1) //cleanup shared by do1d,do2d
end
//getdata: take a point in one data streamfunction getdata(stream)
variable stream //from datatypes panelSVAR datatype=$("S_datatype"+num2istr(stream))NVAR dmm=$("V_dmm"+num2istr(stream)), gtype=$("V_gtype"+num2istr(stream))variable g,i,v,Lsens,Isens,Rinline,ZeroOffset,Vapp,Rbias,DCgain, R0=25812.8variable/G V_hpampsum // AC amplitude of HP33250 pairvariable/G V_gtrip // indicator for too low a conductance
138
variable/G V_limitgflag // check for too low conductance?variable/G V_glimit // limit to check against
if(stringmatch(datatype[0],"g")||(stringmatch(datatype[0,3],"cond")))if(gtype==1) // 1=V-bias, lockin source
Lsens=2e-3Vapp=6e-6Rinline=4.49e4Isens=1e-9ZeroOffset=0i=Isens*Lsens*(readdmm(dmm)-ZeroOffset)/10000g= R0/((Vapp/i)-Rinline)
elseif(gtype==2) // 2=V-bias using HP33250 as sourceLsens=100e-3Rinline=20000Isens=1e-9ZeroOffset=0i=Isens*Lsens*(readdmm(dmm)-ZeroOffset)/10000g= R0/((V_hpampsum/(1e6*i))-Rinline)
elseif(gtype==3) // 3=V-bias, reading a resistor voltage, HP sourceLsens=5e-4Rbias=1e3ZeroOffset=0v=Lsens*(readdmm(dmm)-ZeroOffset)/10000g= R0/(((V_hpampsum/(1000*v))-1)*Rbias-Rinline)
elseif(gtype==4) // 4=I-bias using lockin as sourceLsens=50e-6Vapp=1e-1Rbias=1e8ZeroOffset=0v=Lsens*(readdmm(dmm)-ZeroOffset)/10000g=(R0/Rbias)*((Vapp/v)-1)
elseif(gtype==5) // 5=I-bias using HP33250Lsens=100e-6Rbias=10e6ZeroOffset=0v=Lsens*(readdmm(dmm)-ZeroOffset)/10000g=(R0/Rbias)*((V_hpampsum/342/(1e3*v))-1)
endifif((V_limitgflag)&&(g<V_glimit)) //stop on low conductance?
V_gtrip=1endifreturn g
elseif(stringmatch(datatype[0],"v")) // voltageLsens=10e-6ZeroOffset=0return Lsens*(readdmm(dmm)-ZeroOffset)/10000
elseif(stringmatch(datatype[0],"i")) // currentDCgain=1e11ZeroOffset=0v= (readdmm(dmm)-ZeroOffset)/(DCgain*1000)return v
endifend
139
//setval: generic routine to set physical parametersfunction setval(idstr,value)
string idstrvariable valueif(stringmatch(idstr[0],"c"))
variable num=str2num(idstr[1,2])chanramp(num,value)
elseif(stringmatch(idstr,"bx"))ramp3axis(1,value)wait(52)
elseif(stringmatch(idstr,"by"))ramp3axis(2,value)wait(52)
elseif(stringmatch(idstr,"bz"))ramp3axis(3,value)wait(52)
elseif(stringmatch(idstr,"time")||stringmatch(idstr,"noise"))wait(0)
elseif(stringmatch(idstr,"diag"))//diagonal "detuning" sweep of left and right wallsvariable/G diag_offset,diag_slopesetval("right",value)setval("left",diag_offset+diag_slope*value)
elseif(stringmatch(idstr,"left"))//change the left wall (c6) while updating left listener (c7)//to keep conductance in rangevariable/G c7offset,c7slope,leftcentersetval("c6",value)setval("c7",c7offset-c7slope*(value-leftcenter))
elseif(stringmatch(idstr,"right"))//same with left wallvariable/G c1offset,c1slope,rightcentersetval("c2",value)setval("c1",c1offset-c1slope*(value-rightcenter))
elseprint "ABORTING - couldn’t resolve idstr"return 0
endifend
//getlabel: returns the label for to the given axis from a wave name string.// also fills S_namesegment#, where # is the axis, with its idstr// FOR NEW PARAMETERS: ADD HERE AND IN SETVALfunction/S getlabel(wavestr,axis)
string wavestrvariable axis //0 is data, 1 is x, 2 is y, 3 is zstring labelstr,cropstr=wavestr,idstrvariable nextpos,paramnum//make the wave "datalabels" and populate using addtext//for each sweep parameter or measurement type add a line here of the form:// addtext(dl,"<idstr>;<label text>")//if a shorter name could conflict with a longer one (eg. "b" and "bx") put//the longer one first, otherwise I’ve used alphabetical order for clarity.
140
//See help on Label for escape codes// note that "\\" gives "\" in the actual stringmake/T/o/n=0 $"datalabels"=""wave/T dl=$"datalabels"addtext(dl,"bx;X Magnetic Field (\\umT)")addtext(dl,"by;Y Magnetic Field (\\umT)")addtext(dl,"bz;Z Magnetic Field (\\umT)")addtext(dl,"bmt;Measured Field (\\umT)")addtext(dl,"b;Magnetic Field (\\umT)")addtext(dl,"cond;g (\\ue\\S2\\M/h)")addtext(dl,"color;")addtext(dl,"c;") //this one is special-see belowaddtext(dl,"data;Data point")addtext(dl,"dif_cond;\\u#2conductance derivative (a.u.)")addtext(dl,"dif_res_cond;\\u#2conductance derivative (a.u.)")addtext(dl,"diag;Diagonal parameter (\\umV)")addtext(dl,"freq;Frequency (\\uHz)")addtext(dl,"g;g (\\ue\\S2\\M/h)")addtext(dl,"i;Current (\\uA)")addtext(dl,"noise;Time (\\usec)")addtext(dl,"norm;Normalized Signal")addtext(dl,"par;Parallel Field (\\umT)")addtext(dl,"perp;Perpendicular Field (\\umT)")addtext(dl,"phase;Phase (\\uRadians)")addtext(dl,"Res_cond;delta g(\\ue\\S2\\M/h)")addtext(dl,"time;Time (\\usec)")addtext(dl,"t1;First pulse time (\\uns)")addtext(dl,"t2;Second pulse delay (\\uns)")addtext(dl,"t3;Second pulse time (\\uns)")addtext(dl,"vdc;Calculated DC voltage (\\uV)")addtext(dl,"v;Voltage (\\uV)")
variable i=0,jdo //loop through axes
nextpos=0j=0do //loop through ids and labels in the text wave "datalabels"
if(j==numpnts(dl)) //quit if no match was foundreturn ""
endifidstr=stringfromlist(0,dl[j],";")//does the next id match dl[j]?if(stringmatch(cropstr[0,strlen(idstr)-1],idstr))
//is there a number assoc. with this id?paramnum=str2num(cropstr[strlen(idstr),30])nextpos=strlen(idstr)+(numtype(paramnum) ? 0 :
strlen(num2istr(paramnum)))if(stringmatch(idstr,"c")) //DAC chans are special
labelstr="Chan " + num2istr(paramnum) + " (mV)"if(waveexists($"DAClabel"))
wave/T cnames=$"DAClabel";labelstr+=" - "+cnames[paramnum]endif
else //For other ids, take the label from "datalabels"labelstr=stringfromlist(1,dl[j],";")
141
endifendifj+=1
while(nextpos==0)SVAR ns=$("S_namesegment"+num2istr(i))if(SVAR_exists(ns))
ns=cropstr[0,nextpos-1]endifcropstr=cropstr[nextpos,30] //move on to the next section of the namei+=1
while(i<=axis)return labelstr
end
//alexepsfix.ipf Alex Johnson 2004// This is my way of making high-quality, compact eps files from Igor// When Igor exports an image plot as eps, it makes the interior of the image// be a single object (good) but all pixels at the edges, no matter what you do,// are drawn as rectangles. This has several bad consequences:// 1. big files// 2. they don’t always draw in the same places as the other pixels// 3. they don’t always get the same colors as the other pixels// Igor also draws the entire color scale as rectangles, with the same// annoying results// To fix, first run "fixgraphedges" for each graph to be exported, then// export the layout as a color eps, then run "removerects" and select the// exported file. Finally, open the "..._out.eps" file in Illustrator and// insert the appropriate color scale file (which, if you don’t have it, you can// make by exporting one as a TIFF or similar, cropping to just the colorbar// and resizing to 256 wide by 1 pixel high).
//fixgraphedges: rounds axes so that Igor will make small rectangles around the// edges which will wind up under the axes so they can be thrown awayfunction fixgraphedges(overagex,overagey)
variable overagex,overagey //fraction of a pixel to make the overagewave w=$topimage(0)variable x0=dimoffset(w,0),dx=abs(dimdelta(w,0))variable y0=dimoffset(w,1),dy=abs(dimdelta(w,1))variable V_min,V_maxgetaxis/Q bottom; variable xmin=V_min,xmax=V_maxgetaxis/Q left; variable ymin=V_min,ymax=V_max
if(xmax<xmin) //to account for reversed axesoveragex*=-1
endifif(ymax<ymin)
overagey*=-1endifxmax=(round((xmax-x0)/dx-(overagex+.5))+(overagex+.5))*dx+x0xmin=(round((xmin-x0)/dx+(overagex+.5))-(overagex+.5))*dx+x0ymax=(round((ymax-y0)/dy-(overagey+.5))+(overagey+.5))*dy+y0ymin=(round((ymin-y0)/dy+(overagey+.5))-(overagey+.5))*dy+y0
142
setaxis bottom,xmin,xmaxsetaxis left,ymin,ymax
end
//removerects: takes an EPS file saved by Igor and removes the little rectangles// made by fixgraphedges, saving the result in a new file with "_out" appendedfunction removerects()
variable refnumin,refnumoutstring S_filename,filenameout,instrvariable V_filepos,V_logEOFopen/R/T=".eps"/M="Select EPS file to fix" refnuminif(refnumin==0) //quit if no file selected
return 0endiffilenameout=stringfromlist(0,S_filename,".")+"_out.eps"open refnumout as filenameoutstring nrstr="NewRectImageLine"+num2char(13),cxstr="cx"+num2char(13)do
//read a line from the input file (terminator CR)freadline/T=(num2char(13)) refnumin,instr//include in output only if it doesn’t contain "NewRectImageLine" or "cx"if((strsearch(instr,nrstr,0)<5)&&(strsearch(instr,cxstr,0)<0))
fprintf refnumout,"%s",instr //spit it back out if it’s OKendiffstatus refnumin
while(V_filepos<V_logEOF)close refnumin; close refnumout
end
143
Appendix E
Measurement techniques
This appendix describes some techniques I use for efficient device measurement. I was
expecting to also include here some more detailed information about fabrication of
my devices (which I spent at least an integrated year on), but essentially everything
I did can be found in older Marcus group theses [1, 161] and in fact the group has
significantly moved on already since I stopped doing fab.1 What remains for this
appendix are two parts: the sequence of events (tests and cooling steps) I use to
safely cool a device while learning as early as possible if there is a problem, and ways
I’ve used “wall control,” a slightly outdated term for adjusting gates automatically
for quicker movements around parameter space.
E.1 Cooling and diagnostics
While the fridge is still warm, you can check everything about the fridge wiring
(except whether there are shorts/opens that occur only cold) and quite a bit about
1See the group website (http://marcuslab.harvard.edu/) under “member informa-tion”.
144
18
9
16
17 24
25
32
Figure E.1: Chip carrier for testing fridge wiring. Solid rectangles are bond pads,and the curves and lines represent bond wires connecting them. When inserted asusual, this connects 1–2, 3–4, etc., but when turned 90, it connects all different pairs(1–24, 2–3, etc.).
the device. First, while the device is still safe in its antistatic case, check the fridge
wiring. The easiest way to do this is to have a chip carrier bonded pad-to-pad in a
pattern that links different gates depending on the insertion orientation. The one I
made has bonds as shown in Fig. E.1. Insert this carrier in the normal orientation.
Ground all gates on the breakout box (At this stage it is often easiest to take off all
grounding caps and ground the gates only with the switches. Do NOT do this when
the device is inserted, because inadvertent switch flipping could blow up a gate.
Attach a multimeter in resistance mode from line 1 to ground by attaching a BNC-
to-banana connector to the multimeter, and plug this into line 1 with a BNC cable.
Then float line 1. The ohmmeter should read roughly the inline resistance of line 1
plus that of line 2 (or whichever line 1 is paired with). If it shows overload, one of
these lines is broken. Now float line 2, and the ohmmeter should overload. If there
is any finite resistance showing, one of these lines is shorted to another line or to
ground. Repeat this procedure with each pair of gates. If there are no problems, the
sample lines are fine.2 If you found a problem, repeat those gates after turning the
2There’s still the possibility that lines 1 and 2, for example, are shorted together,so a truly complete check would involve doing this again with the carrier turned 90
145
Ithaco
In Out
to DMM
Battery Box
+ -
Breakout Box
Line(s)to test
Unusedohmics
Otherlines
Figure E.2: Gate/ohmic test circuit. The Ithaco current meter input is connected tothe negative side of the battery, and the positive side of the battery is connected tothe gates or ohmics to be tested. These terminals may be reversed to test current atthe opposite bias. Some ohmics may be floated during ohmic tests, while all othergates and the return-path ohmics should be grounded to prevent shocks.
carrier 90 to isolate the broken or shorted line.
E.1.1 Room-temperature sample tests
Once the fridge lines are working you can insert the sample. The sample is more
sensitive to shocks when cold than warm, because there is more leakage in a warm
sample, but it can still very easily be destroyed, so be sure to ground yourself well to
the fridge (and ground all sample lines) while inserting the sample. All initial tests
should use voltage bias so that a potential open circuit won’t generate huge voltages,
and you should turn the voltage slowly up from zero while watching the current, so
that a potential short circuit won’t generate huge currents. At room temperature, the
exact resistances or I/V characteristics are unimportant, what matters is their order
of magnitude, so I use a simple battery/current meter circuit as shown in Fig. E.2
for all room-temperature tests. In this circuit, current flows from the device, through
the battery, to the current meter, and back again through the BNC shields and shells
anyway.
146
of the battery box and breakout box to the low-voltage side of the device.
First test the ohmics. With all gates grounded, attach the circuit of Fig. E.2 to one
ohmic at a time with all other ohmics grounded. Slowly turn the battery up to several
mV (1–2 hundredths on the dial), watching the current. Ohmic resistances should be
in the neighborhood of 10 kΩ or less at room temperature. If these resistances are
in the MΩ range, they are likely not to work at all at low temperature, but in the
kΩ range, room-temperature resistance may not be correlated with low-temperature
resistance, so it is not worthwhile to measure the pairwise resistances at this point.
If the resistance is in the GΩ range, the bond wire likely popped off or the chip
carrier is not sitting well in its socket. Note that the large, many-channel battery
boxes typically have about 100 kΩ output impedances whereas some of the individual
battery boxes have vanishing output impedance at low voltage, and the Ithaco on 10−9
A/V sensitivity has ∼4 kΩ input impedance.3 The impedance of the measurement
circuit should be determined and taken into account when measuring resistances in
the kΩ range by this method.
Next, test the gates for leakage, which will also tell you whether the bond wires are
connected and if any gates are shorted together. With everything grounded, including
the gate to be tested (remove its grounding cap but keep the switch grounded), turn
the battery to zero and attach it to the gate. You should see the measured current
jump from essentially zero (when it was measuring an open BNC) to somewhere
around the high pA to nA range (depending on the measurement impedance) just
3The manual says 20 kΩ, but the ones I’ve measured appear to all have only 4 kΩ.The difference persists at other scales:
Sensitivity (A/V) 10−11 10−10 10−9 10−8 10−7 10−6
Listed R (Ω) 2M 200k 20k 2k 200 20Measured R (Ω) 390k 42k 4.3k 420 81 <10
147
based on the small residual voltage from the battery, which serves as a confirmation
that the circuit is properly attached and it is safe to unground the gate. Do so, and
you should see the current drop to near the “zero” value it had open. A current of
a few pA or less remaining is normal, this is photocurrent in the device. You can
see this by covering the sample with your hand or with the cold finger can, and the
current will change. Now slowly turn the battery up, watching the current. I usually
go to 300 mV (one turn, either positive or negative will do) and stop if the current
goes above 10 nA (10 V with the Ithaco on 10−9) although occasionally the leakage
will be up to ∼100 nA in the light. Photoconductivity is an even bigger effect than
photocurrent in these devices, and at room temperature in the dark a good gate
should not have more than a few nA leakage current at 300 mV. Also note that if
you are testing a gate connected through a bias tee, it has a large capacitance to fill
through an inductor and a resistor, so it has big transient currents on a timescale of
about a second, and you must proceed slowly, waiting for the transients to settle to
see that the DC current is safe.
If any gate leaks much more than a few nA, it may have a short. If more than one gate
leaks and they are nearby, they may be shorted together. A safe way to test this is to
measure the leakage through one of them while grounding the others through large
resistors (1–10 GΩ, bigger than the leakage resistance). If the leakage is gate-to-gate,
it will drop substantially when doing this. The problem may be a conducting residue
or contaminant on the chip, in which case you can clean it and try again (we have
had cases of conducting methanol, for example), or it may be a lithographic error, in
which case either you can live with those gates always having the same voltage or the
device is trash. This is the end of what you can do at room temperature, so if you
148
have gotten this far with no problems, you are ready to cool.
E.1.2 Cooling with positive bias
The practice of cooling with a positive voltage on the gates, which may have started
in the Marcus group but is now used widely, recently moved beyond the realm of
mythology. Andy Sachrajda’s group did a systematic study (although I don’t believe
this has been published) of switching noise in the conductance of one particular QPC,
and found a clear minimum around +250 mV with more noise on either side. Anec-
dotal evidence in our group suggests that each heterostructure has its own preferred
cooling voltage, and we have typically used anywhere from 250–400 mV. To do this,
leave all of the ohmics grounded and while their grounding caps are still on, flip all
of the gates to bus. Remove one grounding cap and attach the same battery/Ithaco
circuit (making sure you attach it by a long enough BNC cable that you can still get
the fridge into the dewar), then remove the grounding caps from all of the other gates.
Now slowly turn up the voltage to the desired value, watching the current to see that
it’s roughly the sum of all of the individual gate leakages (in the dark, assuming you
have the IVC closed). Now you can start cooling, and the leakage current should
drop sharply,4 but keep the positive bias on until you reach 4 K.
4Ron claims he saw the sample cool just due to gas expansion upon pumping outthe IVC this way.
149
E.1.3 Low-temperature sample tests
By the time you get to 4 K, leakage at +300 mV should be below 1 nA and negative-
bias leakage should be immeasurably small up to several volts. Unless you are using
an exotic heterostructure, there should be no reason to test how far you can go in
negative bias as long as the positive-bias leakage is small. If you do push the gates all
the way to negative-bias leakage, you may make the device noisy again. You can now
take off the positive bias, test the ohmics individually, and test gate action. At this
point the actual numbers are useful and are unlikely to change as you cool further
(although I have heard of cases where ohmics work at 4 K but not at 50 mK) so it is
worthwhile setting up a lockin measurement. The following diagnostics can be done
in parallel with cooling the fridge from 4 K to base. Again, voltage bias is safer for
diagnostics, and the usual configuration is to take the lockin oscillator through a big
divider (usually 105:1) to one ohmic and out another to the Ithaco, with all other
ohmics floated. You can do this from one ohmic to each other ohmic to make sure
the mesa conducts and all of the ohmics basically work, then test individual ohmic
pairs on your current paths. Finally, this same measurement configuration can be
used to test gate action by applying negative voltages to a pair of gates which should
constrict the current path being measured. Slowly turn up the (negative) voltage on
one gate, and you should see a small drop in the conductance when it depletes the
2DEG underneath it. You should see a bigger drop in conductance when the second
gate depletes and current is forced to go through the channel between the two gates.
Depletion should occur around -200–300 mV, and can be used to test gates that are
too far away from each other to pinch off completely. for QPCs, continue turning the
two gates up uniformly and conductance should drop to zero at anywhere from -500
150
mV to -2500 mV, depending on the design and the 2DEG depth and density. When
it does, you should be able to push the channel back and forth by pulling back on one
gate until conduction turns back on, then pushing on the other gate to turn it back off
again. This is the real signature that the QPC gates are working correctly, and you
can check that nearby plungers have an effect on conductance as well. At this point,
if you STILL haven’t encountered any problems, you are ready to put the gates on
DACs, armed with estimates of where pinch-off points are based on the above tests.
E.2 Wall control
Wall control is a slightly anachronistic name the group uses to describe correcting
for the cross-coupling between gates. The technique was originally used to adjust the
walls (the gates next to the QPCs of a dot) to keep a constant number of channels
open in the QPCs while the plungers (the little gates far from the QPCs) were swept
over a wide range to distort the shape and size of the dot. I’ve done several different
analogous operations to keep devices at the appropriate operating points, which I will
describe here in order of increasing complexity.
E.2.1 Linear charge sensor control
The experiments in Chs. 7–9 all used devices where the optimal way to control the dot
electrons was to sweep a gate that also directly affected the charge sensor (gate 2 or 6
in Fig. 7.1) and therefore a sweep of any reasonable size would push the charge sensor
out of its high-sensitivity region. We corrected for this by changing the other gate
151
defining the charge sensor QPC (gate 1 or 7) in the opposite direction, for example:
c1 = c01 − slope(c2 − c02). This was built into the sweep parameter (see “left” in the
routine setval, Ap. D) so that whenever c2 was changed, c1 would change as well.
The offset c02 is taken to be the value of gate 2 at the most interesting point in the
sweep, and c01 is determined by sweeping just c1 at this point and finding the optimal
sensor operating point, usually by picking a conductance value, say 0.3 e2/h. Lastly,
slope is determined by changing c2 either direction as far as you would want to sweep
it and taking new c1 sweeps to see how far the optimal operating point moved.
E.2.2 Automatic charge sensing
From time to time the charge sensor will change its conductance, so that what was
the optimal sensing point at the beginning of a long sweep may no longer be sensitive
at the end. In addition, changes in other gates can have an effect on the charge
sensor as well, but these effects may be nonlinear, or dependent on the values of
other gates, and in any event it would be time-consuming to explicitly correct for
all extra gates. As long as these gates are not being changed within a single data
sweep (or even if they are, so long as they aren’t the inner loop) we can automatically
adjust the sensor by running a feedback routine between sweeps which changes the
sensor setpoint while measuring its conductance until this conductance is optimized.
This can work within the linear charge sensor control framework described above by
simply setting c02 to the value of c2 when the routine is run, and setting c01 to the
value of c1 when the routine terminates. With a routine like this, it is important to
build in error checking so that the feedback routine doesn’t run away to high positive
or negative gate voltages.
152
E.2.3 Honeycomb centering
In double quantum dots, one often wants to pick a honeycomb vertex pair and opti-
mize its characteristics by changing the three tunnel barriers bounding the dots. This
was particularly important in the few-electron dots of Chs. 7–9, because the majority
of data could only be taken at the (1,1)–(0,2) triple-point pair. With such small dots,
the strengths of tunnel barriers are not determined simply by a linear combination
of all gates, so wall control as originally envisioned is not possible. However, it turns
out that the position of the vertices is, to a good approximation, simply a linear
combination of all gate voltages, and we made use of that feature to automatically
center the vertices in two-dimensional sweeps (c2 and c6) as the other gates (c3, c4,
c5 and c12) are adjusted. By measuring the vertex locations at several values of each
of these gates, we can assemble equations of the form
ccenter2 = c02 −K23(c3 − c03)−K24(c4 − c04)− . . .
and
ccenter6 = c06 −K63(c3 − c03)−K64(c4 − c04)− . . .
and it is a simple matter to programmatically center the sweeps at these locations
for any given values of the extra gates.
E.2.4 Matrix wall control
The device from Ch. 4 and other medium-sized but complicated devices are amenable
to a full “exact diagonalization” wall control (to borrow a term that makes the tech-
nique sound more fancy than it is). In this device (see Fig. 4.1) I had five physical
153
parameters I cared about (although this was not necessary for the Fano experiment
described here, only for the ill-fated Coulomb drag experiment which followed it): the
upper left barrier, the lower left barrier, the dot-channel barrier, the channel conduc-
tance, and a plunger. Each parameter had one gate which was primarily responsible
for controlling it, but these gates all affected all of the other parameters as well. In
order to make any sense of this I had to measure all of the cross couplings, put them
into a matrix, and invert the matrix so that I could tell the program what to make
each parameter and it would calculate all of the gate values.
The first step is to define the physical parameters in terms of the primary control
gate. For example, a QPC barrier parameter could be defined as “mV of gate 1 from
a conductance of 0.5 e2/h.”5 Then take a number of sweeps of gate 1 at various values
of gate 2 and measure how far the center point (i.e. 0.5 e2/h) moves. If, for example,
every time you add 100 mV to c2 it moves you back 21 mV in c1 and adding 100 mV to
c3 moves you back 7 mV in c1 and so on, you can construct an equation (similar to the
procedure in the previous section) like p1 = c1 +0.21c2 +0.07c3 + . . . (I’ve suppressed
offsets in ci, and will add them back later). This procedure can be repeated for all of
the barrier parameters, for the plunger you could attempt to do wall control on dot
size (say, by taking the slope of Coulomb blockade peaks or conductance fluctuations
in 2D plots) but in my case I was interested only in statistics as a function of dot size,
not following individual features, so if gate 4 was the plunger I simply set p4 = c4.
After all of this you have a matrix equation p = M(c − c0) where c0 is the set
5Notice that this kind of wall control can’t be linearized about a point with balancedleads, because we need to open another QPC up to one or two full modes in orderto independently measure the first QPC, but for all but the smallest devices thenonlinearities should be small enough that this doesn’t matter.
154
of gate voltages you linearized about. Now you can invert this equation,6 yielding
c = c0 +M−1p. Now if you make a wave to store the values of the parameters pi you
can use them as sweep parameters, plugging them into this matrix equation every
time you change one and using the output to set the gate voltages. These parameters
have roughly the same scaling as the original gate voltages (so a 1-mV change in p1
will correspond to approximately a 1-mV change in c1 with associated corrections to
the other gates) but they are centered about zero unless you specifically set them
otherwise.
6Igor version 5 now has a MatrixInverse command to get this directly, in priorversions you had to get it as a byproduct of MatrixGaussJ, whatever that means.
155
Bibliography
[1] Cronenwett, S. M. Coherence, charging, and spin effects in quantum dots andpoint contacts. PhD thesis, Stanford University, (2001).
[2] Mabuchi, H. and Doherty, A. C. Cavity quantum electrodynamics: Coherencein context. Science 298, 1372–7 (2002).
[3] Keller, M., Lange, B., Hayasaka, K., Lange, W., and Walther, H. Continuousgeneration of single photons with controlled waveform in an ion-trap cavitysystem. Nature 431, 1075–8 (2004).
[4] Vahala, K. J. Optical microcavities. Nature 424, 839–46 (2003).
[5] Badolato, A., Hennessy, K., Atature, M., Dreiser, J., Hu, E., Petroff, P. M.,and Imamoglu, A. Deterministic coupling of single quantum dots to singlenanocavity modes. Science 308, 1158–61 (2005).
[6] Jarillo-Herrero, P., Sapmaz, S., Dekker, C., Kouwenhoven, L. P., and van derZant, H. S. J. Electron-hole symmetry in a semiconducting carbon nanotubequantum dot. Nature 429, 389–92 (2004).
[7] Mason, N., Biercuk, M. J., and Marcus, C. M. Local gate control of a carbonnanotube double quantum dot. Science 303, 655–8 (2004).
[8] Borgstrom, M. T., Zwiller, V., Muller, E., and Imamoglu, A. Optically brightquantum dots in single nanowires. Nano Lett. 5, 1439–43 (2005).
[9] Davidovic, D. and Tinkham, M. Spectroscopy, interactions, and level splittingsin Au nanoparticles. Phys. Rev. Lett. 83, 1644–7 (1999).
[10] Petta, J. R. and Ralph, D. C. Measurements of strongly anisotropic g factorsfor spins in single quantum states. Phys. Rev. Lett. 89, 156802 (2002).
[11] Kroutvar, M., Ducommun, Y., Heiss, D., Bichler, M., Schuh, D., Abstreiter,G., and Finley, J. J. Optically programmable electron spin memory using semi-conductor quantum dots. Nature 432, 81–84 (2004).
156
[12] Bracker, A. S., Stinaff, E. A., Gammon, D., Ware, M. E., Tischler, J. G.,Shabaev, A., Efros, A. L., Park, D., Gershoni, D., Korenev, V. L., andMerkulov, I. A. Optical pumping of the electronic and nuclear spin of singlecharge-tunable quantum dots. Phys. Rev. Lett. 94, 047402 (2005).
[13] Chiorescu, I., Bertet, P., Semba, K., Nakamura, Y., Harmans, C. J. P. M., andMooij, J. E. Coherent dynamics of a flux qubit coupled to a harmonic oscillator.Nature 431, 159–62 (2004).
[14] Wallraff, A., Schuster, D. I., Blais, A., Frunzio, L., Huang, R. S., Majer, J., Ku-mar, S., Girvin, S. M., and Schoelkopf, R. J. Strong coupling of a single photonto a superconducting qubit using circuit quantum electrodynamics. Nature 431,162–7 (2004).
[15] Stormer, H. L., Chang, A., Tsui, D. C., Hwang, J. C. M., Gossard, A. C., andWeigmann, W. Fractional quantization of the Hall-effect. Phys. Rev. Lett. 50,1953–6 (1983).
[16] Stormer, H. L., Tsui, D. C., and Gossard, A. C. The fractional quantum Halleffect. Rev. Mod. Phys. 71, S298–305 (1999).
[17] Eisenstein, J. P. and MacDonald, A. H. Bose-einstein condensation of excitonsin bilayer electron systems. Nature 432, 691–4 (2004).
[18] Thornton, T. J., Pepper, M., Ahmed, H., Andrews, D., and Davies, G. J.One-dimensional conduction in the 2D electron-gas of a GaAs/AlGaAs hetero-junction. Phys. Rev. Lett. 56, 1198–1201 (1986).
[19] van Wees, B. J., van Houten, H., Beenakker, C. W. J., Williamson, J. G.,Kouwenhoven, L. P., van der Marel, D., and Foxon, C. T. Quantized conduc-tance of point contacts in a two-dimensional electron gas. Phys. Rev. Lett. 60,848–91 (1988).
[20] Cronenwett, S. M., Lynch, H. J., Goldhaber-Gordon, D., Kouwenhoven, L. P.,Marcus, C. M., Hirose, K., Wingreen, N. S., and Umansky, V. Low-temperaturefate of the 0.7 structure in a point contact: A Kondo-like correlated state in anopen system. Phys. Rev. Lett. 88, 226805 (2002).
[21] Auslaender, O. M., Steinberg, H., Yacoby, A., Tserkovnyak, Y., Halperin, B. I.,Baldwin, K. W., Pfeiffer, L. N., and West, K. W. Spin-charge separation andlocalization in one dimension. Science 308, 88–92 (2005).
[22] van Wees, B. J., Kouwenhoven, L. P., Harmans, C. J. P. M., Williamson, J. G.,Timmering, C. E., Brockaart, M. E. I., Foxon, C. T., and Harris, J. J. Obser-vation of zero-diomensional states in a one-dimensional electron interferometer.Phys. Rev. Lett. 62, 2523–6 (1989).
157
[23] Meirav, U., Kastner, M. A., and Wind, S. J. Single-electron charging andperiodic conductance resonances in GaAs nanostructures. Phys. Rev. Lett. 65,771–4 (1990).
[24] Kastner, M. A. The single-electron transistor. Rev. Mod. Phys. 64, 849–58(1992).
[25] Kouwenhoven, L. P., Austing, D. G., and Tarucha, S. Few-electron quantumdots. Rep. Prog. Phys. 64, 701–36 (2001).
[26] Drummond, T. J., Masselink, W. T., and Morkoc, H. Modulation-doped GaAs/-(Al,Ga)As heterojunction field-effect transistors—MODFETs. Proc. IEEE 74,773–822 (1986).
[27] Kristensen, A., Bruus, H., Hansen, A. E., Jensen, J. B., Lindelof, P. E., Marck-mann, C. J., Nygard, J., and Sørensen, C. B. Bias and temperature dependenceof the 0.7 conductance anomaly in quantum point contacts. Phys. Rev. B 62,10950–7 (2000).
[28] Fuhrer, A., Luscher, S., Ihn, T., Heinzel, T., Ensslin, K., Wegscheider, W., andBichler, M. Energy spectra of quantum rings. Nature 413, 822–5 (2001).
[29] Fuhrer, A., Dorn, A., Luscher, S., Heinzel, T., Ensslin, K., Wegscheider, W.,and Bichler, M. Electronic properties of nanostructures defined in Ga[Al]Asheterostructures by local oxidation. Superlattices and Microstructures 31, 19–42 (2002).
[30] Crook, R., Graham, A. C., Smith, C. G., Farrer, I., Beere, H. E., and Ritchie,D. A. Erasable electrostatic lithography for quantum components. Nature 424,591–4 (2003).
[31] Zheng, H. Z., Wei, H. P., Tsui, D. C., and Weimann, G. Gate-controlledtransport in narrow GaAs/AlxGa1−xAs heterostructures. Phys. Rev. B 34,5635–8 (1986).
[32] Stopa, M. Quantum dot self-consistent electronic structure and the Coulombblockade. Phys. Rev. B 54, 13767–83 (1996).
[33] Stopa, M., van der Wiel, W. G., Franceschi, S. D., Tarucha, S., and Kouwen-hoven, L. P. Magnetically induced chessboard pattern in the conductance of akondo quantum dot. Phys. Rev. Lett. 91, 046601 (2003).
[34] Hatano, T., Stopa, M., and Tarucha, S. Single-electron delocalization in hybridvertical-lateral double quantum dots. Science 309, 268–71 (2005).
[35] Topinka, M. A., LeRoy, B. J., Westervelt, R. M., Shaw, S. E. J., Fleischmann,R., Heller, E. J., Maranowski, K. D., and Gossard, A. C. Coherent branchedflow in a two-dimensional electron gas. Nature 410, 183–6 (2001).
158
[36] Vanicek, J. and Heller, E. J. Uniform semiclassical wave function for coherenttwo-dimensional electron flow. Phys. Rev. E 67, 016211 (2003).
[37] Datta, S. Electronic Transport in Semiconductor Systems. Cambridge Univer-sity Press, (1997).
[38] Imry, Y. Introduction to Mesoscopic Physics. Oxford University Press, (1997).
[39] Kouwenhoven, L. P., Marcus, C. M., McEuen, P. L., Tarucha, S., Westervelt,R. M., and Wingreen, N. S. Electron transport in quantum dots. In MesoscopicElectron Transport. Kluwer (1997).
[40] van der Wiel, W. G., Franceschi, S. D., Elzerman, J. M., Fujisawa, T., Tarucha,S., and Kouwenhoven, L. P. Electron transport through double quantum dots.Rev. Mod. Phys. 75, 1–22 (2003).
[41] Kane, B. E., Facer, G. R., Dzurak, A. S., Lumpkin, N. E., Clark, R. G., Pfeiffer,L. N., and West, K. W. Quantized conductance in quantum wires with gate-controlled width and electron density. Appl. Phys. Lett. 72, 3506–8 (1998).
[42] de Picciotto, R., Pfeiffer, L. N., Baldwin, K. W., and West, K. W. Nonlinearresponse of a clean one-dimensional wire. Phys. Rev. Lett. 92, 036805 (2004).
[43] Cronenwett, S. M., Maurer, S. M., Patel, S. R., Marcus, C. M., Duruoz, C. I.,and Harris, J. S. Mesoscopic Coulomb blockade in an open-channel quantumdot. Phys. Rev. Lett. 81, 5904–7 (1998).
[44] Marcus, C. M., Westervelt, R. M., Hopkins, P. F., and Gossard, A. C. Phasebreaking in ballistic quantum dots—experiment and analysis based on chaoticscattering. Phys. Rev. B 48, 2460–4 (1993).
[45] Bird, J. P., Ishibashi, K., Ferry, D. K., Ochiai, Y., Aoyagi, Y., and Sugano,T. Phase breaking in ballistic quantum dots—transition from 2-dimensional tozero-dimensional behavior. Phys. Rev. B 51, 18037–40 (1995).
[46] Liang, C. T., Castleton, I. M., Frost, J. E. F., Barnes, C. H. W., Smith, C. G.,Ford, C. J. B., Ritchie, D. A., and Pepper, M. Resonant transmission throughan open quantum dot. Phys. Rev. B 55, 6723–6 (1997).
[47] Zumbuhl, D. M., Miller, J. B., Marcus, C. M., Campman, K., and Gossard,A. C. Spin-orbit coupling, antilocalization, and parallel magnetic fields in quan-tum dots. Phys. Rev. Lett. 89, 276803 (2002).
[48] Folk, J. A., Potok, R. M., Marcus, C. M., and Umansky, V. A gate-controlledbidirectional spin filter using quantum coherence. Science 299, 679–82 (2003).
[49] Tarucha, S., Austing, D. G., Honda, T., van der Hage, R. J., and Kouwenhoven,L. P. Shell filling and spin effects in a few electron quantum dot. Phys. Rev.Lett. 77, 3613–6 (1996).
159
[50] Stewart, D. R. Level spectroscopy of a quantum dot. PhD thesis, StanfordUniversity, (1999).
[51] Franceschi, S. D., Hanson, R., van der Wiel, W. G., Elzerman, J. M., Wijpkema,J. J., Fujisawa, T., Tarucha, S., and Kouwenhoven, L. P. Out-of-equilibriumKondo effect in a mesoscopic device. Phys. Rev. Lett. 89, 156801 (2002).
[52] Patel, S. R., Cronenwett, S. C., Stewart, D. R., Huibers, A. G., Marcus, C. M.,Duruoz, C. I., Harris, J. S., Campman, K., and Gossard, A. C. Statistics ofCoulomb blockade peak spacings. Phys. Rev. Lett. 80, 4522–5 (1998).
[53] Sachrajda, A. S., Hawrylak, P., Ciorga, M., Gould, C., and Zawadzki, P. Spinpolarized injection into a quantum dot by means of the spatial separation ofspins. Physica E 10, 493–8 (2001).
[54] Pioro-Ladriere, M., Ciorga, M., Lapointe, J., Zawadzki, P., Korkusinski, M.,Hawrylak, P., and Sachrajda, A. S. Spin-blockade spectroscopy of a two-levelartificial molecule. Phys. Rev. Lett. 91, 026803 (2003).
[55] Elzerman, J. M., Hanson, R., van Beveren, L. H. W., Witkamp, B., Vander-sypen, L. M. K., and Kouwenhoven, L. P. Single-shot read-out of an individualelectron spin in a quantum dot. Nature 430, 431–5 (2004).
[56] Biercuk, M. J. Local gate control in carbon nanotube quantum devices. PhDthesis, Harvard University, (2005).
[57] Chan, I. H., Fallahi, P., Westervelt, R. M., Maranowski, K. D., and Gossard,A. C. Capacitively coupled quantum dots as a single-electron switch. PhysicaE 17, 584–8 (2003).
[58] Lu, W., Ji, Z. Q., Pfeiffer, L., West, K. W., and Rimberg, A. J. Real-timedetection of electron tunnelling in a quantum dot. Nature 423, 422–5 (2003).
[59] Johnson, A. C., Marcus, C. M., Hanson, M. P., and Gossard, A. C. Coulomb-modified Fano resonance in a one-lead quantum dot. Phys. Rev. Lett. 93, 106803(2004).
[60] Fano, U. Effects of configuration interaction on intensities and phase shifts.Phys. Rev. 124, 1866–78 (1961).
[61] Madhavan, V., Chen, W., Jamneala, T., Crommie, M. F., and Wingreen, N. S.Tunneling into a single magnetic atom: Spectroscopic evidence of the Kondoresonance. Science 280, 567–9 (1998).
[62] Gores, J., Goldhaber-Gordon, D., Heemeyer, S., Kastner, M. A., Shtrikman, H.,Mahalu, D., and Meirav, U. Fano resonances in electronic transport through asingle-electron transistor. Phys. Rev. B 62, 2188–94 (2000).
160
[63] Zacharia, I. G., Goldhaber-Gordon, D., Granger, G., Kastner, M. A., Khavin,Y. B., Shtrikman, H., Mahalu, D., and Meirav, U. Temperature dependence ofFano line shapes in a weakly coupled single-electron transistor. Phys. Rev. B64, 155311 (2001).
[64] Fuhner, C., Keyser, U. F., Haug, R. J., Reuter, D., and Wieck, A. D. Phasemeasurements using two-channel Fano interference in a semiconductor quantumdot. Preprint at http://arxiv.org/cond-mat/0307590 (2003).
[65] Kobayashi, K., Aikawa, H., Katsumoto, S., and Iye, Y. Tuning of the Fanoeffect through a quantum dot in an Ahaonov-Bohm interferometer. Phys. Rev.Lett. 88, 256806 (2002).
[66] Kobayashi, K., Aikawa, H., Sano, A., Katsumoto, S., and Iye, Y. Fano resonancein a quantum wire with a side-coupled quantum dot. Phys. Rev. B 70, 035319(2004).
[67] Kim, J., Kim, J.-R., Lee, J.-O., Park, J. W., So, H. M., Kim, N., Kang, K.,Yoo, K.-H., and Kim, J.-J. Fano resonance in crossed carbon nanotubes. Phys.Rev. Lett. 90, 166403 (2003).
[68] Clerk, A. A., Waintal, X., and Brouwer, P. W. Fano resonances as a probe ofphase coherence in quantum dots. Phys. Rev. Lett. 86, 4636–9 (2001).
[69] Xiong, Y.-J. and Xiong, S.-J. Asymmetric line shape and Fano interference inthe transport of electrons through a multilevel quantum dot in the Coulombblockade regime. Phys. Rev. B 66, 153315 (2002).
[70] Song, J. F., Ochiai, Y., and Bird, J. P. Fano resonances in open quantum dotsand their application as spin filters. Appl. Phys. Lett. 82, 4561–3 (2003).
[71] Field, M., Smith, C. G., Pepper, M., Ritchie, D. A., Frost, J. E. F., Jones, G.A. C., and Hasko, D. G. Measurement of Coulomb blockade with a noninvasivevoltage probe. Phys. Rev. Lett. 70, 1311–4 (1993).
[72] Amlani, I., Orlov, A. O., Snider, G. L., Lent, C. S., and Bernstein, G. H.External charge state detection of a double-dot system. Appl. Phys. Lett. 71,1730–2 (1997).
[73] Duncan, D. S., Livermore, C., Westervelt, R. M., Maranowski, K. D., andGossard, A. C. Direct measurement of the destruction of charge quantizationin a single-electron box. Appl. Phys. Lett. 74, 1045–7 (1999).
[74] Chan, I. H., Westervelt, R. M., Maranowski, K. D., and Gossard, A. C. Stronglycapacitively coupled quantum dots. Appl. Phys. Lett. 80, 1818–20 (2002).
[75] Breit, G. and Wigner, E. Capture of slow neutrons. Phys. Rev. 49, 519 (1936).
161
[76] Aharony, A., Entin-Wohlman, O., Halperin, B. I., and Imry, Y. Phase mea-surement in the mesoscopic Aharonov-Bohm interferometer. Phys. Rev. B 66,115311 (2002).
[77] Ng, T. K. and Lee, P. A. On-site Coulomb repulsion and resonant tunneling.Phys. Rev. Lett. 61, 1768–71 (1988).
[78] Taniguchi, T. and Buttiker, M. Friedel phases and phases of transmission am-plitudes in quantum scattering systems. Phys. Rev. B 60, 13814–23 (1999).
[79] Yacoby, A., Heiblum, M., Mahalu, D., and Shtrikman, H. Coherence and phasesensitive measurements in a quantum dot. Phys. Rev. Lett. 74, 4047–50 (1995).
[80] Schuster, R., Buks, E., Heiblum, M., Mahalu, D., Umansky, V., and Shtrik-man, H. Phase measurement in a quantum dot via a double-slit interferenceexperiment. Nature 385, 417–20 (1997).
[81] Baltin, R. and Gefen, Y. An approximate sign sum rule for the transmissionamplitude through a quantum dot. Phys. Rev. Lett. 83, 5094–7 (1999).
[82] Yeyati, A. L. and Buttiker, M. Scattering phases in quantum dots: an analysisbased on lattice models. Phys. Rev. B 62, 7307–15 (2000).
[83] Clerk, A. A. private communication, (2003).
[84] Dicarlo, L., Lynch, H. J., Johnson, A. C., Childress, L. I., Crockett, K., Marcus,C. M., Hanson, M. P., and Gossard, A. C. Differential charge sensing and chargedelocalization in a tunable double quantum dot. Phys. Rev. Lett. 92, 226801(2004).
[85] Gardelis, S., Smith, C. G., Cooper, J., Ritchie, D. A., Linfield, E. H., Jin,Y., and Pepper, M. Dephasing in an isolated double-quantum-dot system de-duced from single-electron polarization measurements. Phys. Rev. B 67, 073302(2003).
[86] Engel, H.-A., Golovach, V. N., Loss, D., Vandersypen, L. M. K., Elzerman,J. M., Hanson, R., and Kouwenhoven, L. P. Measurement efficiency and n-shotreadout of spin qubits. Phys. Rev. Lett. 93, 106804 (2004).
[87] Buks, E., Schuster, R., Heiblum, M., Mahalu, D., and Umansky, V. Dephasingin an electron interference by a ‘which-path’ detector. Nature 391, 871–4 (1998).
[88] Aleiner, I. L., Wingreen, N. S., and Meir, Y. Dephasing and the orthogonalitycatastrophy in tunneling through a quantum dot: the ‘Which Path?’ interfer-ometer. Phys. Rev. Lett. 79, 3740–3 (1997).
[89] Buehler, T. M., Reilly, D. J., Brenner, R., Hamilton, A. R., Dzurak, A. S., andClark, R. G. Correlated charge detection for readout of a solid-state quantumcomputer. Appl. Phys. Lett. 82, 577–9 (2003).
162
[90] Schoelkopf, R. J., Wahlgren, P., Kozhevnikov, A. A., Delsing, P., and Prober,D. E. The radio-frequency single-electron transistor (RF-SET): a fast and ul-trasensitive electrometer. Science 280, 1238–42 (1998).
[91] Elzerman, J. M., Hanson, R., Greidanus, J. S., van Beveren, L. H. W.,Franceschi, S. D., Vandersypen, L. M. K., Tarucha, S., and Kouwenhoven,L. P. Few-electron quantum dot circuit with integrated charge read out. Phys.Rev. B 67, R161308 (2003).
[92] Kouwenhoven, L. P., Oosterkamp, T. H., Danoesastro, M. W. S., Eto, M.,Austing, D. G., Honda, T., and Tarucha, S. Excitation spectra of circular,few-electron quantum dots. Science 278, 1788–92 (1997).
[93] Pothier, H., Lafarge, P., Urbina, C., Esteve, D., and Devoret, M. H. Single-electron pump based on charging effects. Europhys. Lett. 17, 249–54 (1992).
[94] Blick, R. H., Haug, R. J., Weis, J., Pfannkuche, D., von Klitzing, K., andEberl, K. Single-electron tunneling through a double quantum dot: the artificialmolecule. Phys. Rev. B 53, 7899–902 (1996).
[95] Livermore, C., Crouch, C. H., Westervelt, R. M., Campman, K. L., and Gossard,A. C. The Coulomb blockade in coupled quantum dots. Science 274, 1332–5(1996).
[96] Cohen-Tannoudji, C. Quantum Mechanics, volume 1, chapter 4. Wiley (1977).
[97] Ziegler, R., Bruder, C., and Schoeller, H. Transport through double quantumdots. Phys. Rev. B 62, 1961–70 (2000).
[98] Johnson, A. C., Marcus, C. M., Hanson, M. P., and Gossard, A. C. Chargesensing of excited states in an isolated double quantum dot. Phys. Rev. B 71,115333 (2005).
[99] Folk, J. A., Marcus, C. M., and Harris, J. S. Decoherence in nearly isolatedquantum dots. Phys. Rev. Lett. 87, 206802 (2001).
[100] Aleiner, I. L., Brouwer, P. W., and Glazman, L. I. Quantum effects in Coulombblockade. Phys. Rep. 358, 309–440 (2002).
[101] Loss, D. and DiVincenzo, D. P. Quantum computation with quantum dots.Phys. Rev. A 57, 120–6 (1998).
[102] Awschalom, D. D., Loss, D., and Samarth, N. Semiconductor Spintronics andQuantum Computation. Springer, (2002).
[103] Gurvitz, S. A. Measurements with a noninvasive detector and dephasing mech-anism. Phys. Rev. B 56, 15213–23 (1997).
[104] Korotkov, A. N. and Averin, D. V. Continuous weak measurement of quantumcoherent oscillations. Phys. Rev. B 64, 165310 (2001).
163
[105] Pilgram, S. and Buttiker, M. Efficiency of mesoscopic detectors. Phys. Rev.Lett. 89, 200401 (2002).
[106] Hartmann, U. and Wilhelm, F. K. Decoherence of charge states in a doublequantum dot due to cotunneling. Phys. Status Solidi B 233, 385–90 (2002).
[107] Johnson, A. T., Kouwenhoven, L. P., de Jong, W., van der Vaart, N. C., Har-mans, C. J. P. M., and Foxon, C. T. Zero-dimensional states and single-electroncharging in quantum dots. Phys. Rev. Lett. 69, 1592–5 (1992).
[108] Stewart, D. R., Sprinzak, D., Marcus, C. M., Duruoz, C. I., and Harris, J. S.Correlations between ground and excited state spectra of a quantum dot. Sci-ence 278, 1784–8 (1997).
[109] van der Vaart, N. C., de Ruyter van Steveninck, M. P., Kouwenhoven, L. P.,Johnson, A. T., Nazarov, Y. V., Harmans, C. J. P. M., and Foxons, C. T. Time-resolved tunneling of single electrons between Landau levels in a quantum dot.Phys. Rev. Lett. 73, 320–3 (1995).
[110] Sivan, U., Milliken, F. P., Milrove, K., Rishton, S., Lee, Y., Hong, J. M., Boegli,V., Kern, D., and Defranza, M. Spectroscopy, electron-electron interaction, andlevel statistics in a disordered quantum dot. Europhys. Lett. 25, 605–11 (1994).
[111] Fujisawa, T., Austing, D. G., Tokura, Y., Hirayama, Y., and Tarucha, S. Elec-trical pulse measurement, inelastic relaxation, and non-equilibrium transportin a quantum dot. J. Phys. Cond. Mat. 15, R1395–428 (2003).
[112] Elzerman, J. M., Hanson, R., van Beveren, L. H. W., Vandersypen, L. M. K.,and Kouwenhoven, L. P. Excited-state spectroscopy on a nearly closed quantumdot via charge detection. Appl. Phys. Lett. 84, 4617–9 (2004).
[113] Johnson, A. C., Petta, J. R., Marcus, C. M., Hanson, M. P., and Gossard,A. C. Singlet-triplet spin blockade and charge sensing in a few-electron doublequantum dot. Preprint at http://arxiv.org/cond-mat/0410679 (2004).
[114] Chan, I. H., Fallahi, P., Vidan, A., Westervelt, R. M., Hanson, M. P., andGossard, A. C. Few-electron double quantum dots. Nanotechnology 15, 609–13(2004).
[115] Ciorga, M., Sachrajda, A. S., Hawrylak, P., Gould, C., Zawadzki, P., Jullian, S.,Feng, Y., and Wasilewski, Z. Addition spectrum of a lateral dot from Coulomband spin-blockade spectroscopy. Phys. Rev. B 61, R16315–8 (2000).
[116] Kogan, A., Granger, G., Kastner, M. A., Goldhaber-Gordon, D., and Shtrik-man, H. Singlet-triplet transition in a single-electron transistor at zero magneticfield. Phys. Rev. B 67, 113309 (2003).
164
[117] Ono, K., Austing, D. G., Tokura, Y., and Tarucha, S. Current rectification byPauli exclusion in a weakly coupled double quantum dot system. Science 297,1313–7 (2002).
[118] Ashcroft, N. W. and Mermin, N. D. Solid State Physics, chapter 32. Harcourt(1976).
[119] Weinmann, D., Hausler, W., and Kramer, B. Spin blockades in linear andnonlinear transport through quantum dots. Phys. Rev. Lett. 74, 984–7 (1995).
[120] Imamura, H., Aoki, H., and Maksym, P. A. Spin blockade in single and doublequantum dots in magnetic fields: a correlation effect. Phys. Rev. B 57, R4257–60 (1998).
[121] Tokura, Y., Austing, D. G., and Tarucha, S. Single-electron tunneling in twovertically coupled quantum dots. J. Phys. Cond. Mat. 11, 6023–34 (1999).
[122] Ciorga, M., Pioro-Ladriere, M., Zawadzki, P., Hawrylak, P., and Sachrajda,A. S. Tunable negative differential resistance controlled by spin blockade insingle-electron transistors. Appl. Phys. Lett. 80, 2177–9 (2002).
[123] Rokhinson, L. P., Guo, L. J., Chou, S. Y., and Tsui, D. C. Spintronics with Siquantum dots. Microelectronic Engineering 63, 147–53 (2002).
[124] Huttel, A. K., Qin, H., Holleitner, A. W., Blick, R. H., Neumaier, K., Wein-mann, D., Eberl, K., and Kotthaus, J. P. Spin blockade in ground state reso-nance of a quantum dot. Europhys. Lett. 62, 712–8 (2003).
[125] Rogge, M. C., Fuhner, C., Keyser, U. F., and Haug, R. J. Spin blockade incapacitively coupled quantum dots. Appl. Phys. Lett. 85, 606–8 (2004).
[126] Petta, J. R., Johnson, A. C., Marcus, C. M., Hanson, M. P., and Gossard, A. C.Manipulation of a single charge in a double quantum dot. Phys. Rev. Lett. 93,186802 (2004).
[127] Fujisawa, T., Oosterkamp, T. H., van der Wiel, W. G., Broer, B. W., Aguado,R., Tarucha, S., and Kouwenhoven, L. P. Spontaneous emission spectrum indouble quantum dot devices. Science 282, 932–5 (1998).
[128] Koppens, F. H. L., Folk, J. A., Elzerman, J. M., Hanson, R., van Beveren,L. H. W., Fink, I. T., Tranitz, H. P., Wegscheider, W., Kouwenhoven, L. P.,and Vandersypen, L. M. K. Control and detection of singlet-triplet mixing in arandom nuclear field. Science 309, 1346–50 (2005).
[129] Su, B., Goldman, V. J., and Cunningham, J. E. Single-electron tunneling innanometer-scale double-barrier heterostructure devices. Phys. Rev. B 46, 7644–55 (1992).
165
[130] Ashoori, R. C., Stormer, H. L., Weiner, J. S., Pfeiffer, L. N., Baldwin, K. W.,and West, K. W. N-electron ground state energies of a quantum dot in amagnetic field. Phys. Rev. Lett. 71, 613–6 (1993).
[131] Fujisawa, T., Austing, D. G., Tokura, Y., Hirayama, Y., and Tarucha, S. Al-lowed and forbidden transitions in artificial hydrogen and helium atoms. Nature419, 278–81 (2002).
[132] Zumbuhl, D. M., Marcus, C. M., Hanson, M. P., and Gossard, A. C. Cotun-neling spectroscopy in few-electron quantum dots. Phys. Rev. Lett. 93, 256801(2004).
[133] Kyriadikis, J., Pioro-Ladriere, M., Ciorga, M., Sachrajda, A. S., and Hawrylak,P. Voltage-tunable singlet-triplet transition in lateral quantum dots. Phys. Rev.B 66, 035320 (2002).
[134] Wagner, M., Merkt, U., and Chaplik, A. V. Spin-singlet spin-triplet oscillationsin quantum dots. Phys. Rev. B 45, 1951–4 (1992).
[135] Pfannkuche, D., Gudmundsson, V., and Maksym, P. A. Comparison of aHartree, a Hartree-Fock, and an exact treatment of quantum-dot helium. Phys.Rev. B 47, 2244–50 (1993).
[136] Hawrylak, P. Single-electron capacitance spectroscopy of few-electron artificialatoms in a magnetic field—theory and experiment. Phys. Rev. Lett. 71, 3347–50(1993).
[137] van Beveren, L. H. W., Hanson, R., Vink, I. T., Koppens, F. H. L.,Kouwenhoven, L. P., and Vandersypen, L. M. K. Spin filling ofa quantum dot derived from excited-state spectroscopy. Preprint athttp://arxiv.org/cond-mat/0505486 (2005).
[138] Khaetskii, A. V. and Nazarov, Y. V. Spin relaxation in semiconductor quantumdots. Phys. Rev. B 61, 12639–42 (2000).
[139] Erlingsson, S. I., Nazarov, Y. V., and Fal’ko, V. I. Nucleus-mediated spin-fliptransitions in GaAs quantum dots. Phys. Rev. B 64, 195306 (2001).
[140] Khaetskii, A. V., Loss, D., and Glazman, L. Electron spin decoherence inquantum dots due to interaction with nuclei. Phys. Rev. Lett. 88, 186802 (2002).
[141] Merkulov, I. A., Efros, A. L., and Rosen, M. Electron spin relaxation by nucleiin semiconductor quantum dots. Phys. Rev. B 65, 205309 (2002).
[142] de Sousa, R. and Sarma, S. D. Theory of nuclear-induced spectral diffusion:spin decoherence of phosphorus donors in Si and GaAs quantum dots. Phys.Rev. B 68, 115322 (2003).
166
[143] Coish, W. A. and Loss, D. Hyperfine interaction in a quantum dot: non-Markovian electron spin dynamics. Phys. Rev. B 70, 195340 (2004).
[144] Golovach, V. N., Khaetskii, A. V., and Loss, D. Phonon-induced decay of theelectron spin in quantum dots. Phys. Rev. Lett. 93, 016601 (2004).
[145] Johnson, A. C., Petta, J. R., Taylor, J. M., Yacoby, A., Lukin, M. D., Marcus,C. M., Hanson, M. P., and Gossard, A. C. Triplet-singlet spin relaxation vianuclei in a double quantum dot. Nature 435, 925–8 (2005).
[146] Petta, J. R., Johnson, A. C., Yacoby, A., Marcus, C. M., Hanson,M. P., and Gossard, A. C. Pulsed-gate measurements of the singlet-triplet relaxation time in a two-electron double quantum dot. Preprint athttp://arxiv.org/cond-mat/0412048 (2004).
[147] Wald, K. R., Kouwenhoven, L. P., McEuen, P. L., van der Vaart, N. C., andFoxon, C. T. Local dynamic nuclear polarization using quantum point contacts.Phys. Rev. Lett. 73, 1011–4 (1994).
[148] Salis, G. Optical manipulation of nuclear spin by a two-dimensional electrongas. Phys. Rev. Lett. 86, 2677–80 (2001).
[149] Kumada, N., Muraki, K., Hashimoto, K., and Hirayama, Y. Spin degree of free-dom in the ν=1 bilayer electron system investigated via nuclear spin relaxation.Phys. Rev. Lett. 94, 096802 (2005).
[150] Smet, J. H., Deutchmann, R. A., Ertl, F., Wegscheider, W., Abstreiter, G., andvon Klitzing, K. Gate-voltage control of spin interactions between electrons andnuclei in a semiconductor. Nature 415, 281–6 (2002).
[151] Ono, K. and Tarucha, S. Nuclear-spin-induced oscillatory current in spin-blockaded quantum dots. Phys. Rev. Lett. 92, 256803 (2004).
[152] Paget, D., Lampel, G., Sapoval, G., and Safarov, V. I. Low field electron-nuclearspin coupling in gallium-arsenide under optical-pumping conditions. Phys. Rev.B 15, 5780–96 (1977).
[153] Dobers, M., von Klitzing, K., Schneider, J., Weimann, G., and Ploog, K. Electri-cal detection of nuclear magnetic-resonance in GaAs-AlxGa1−xAs heterostruc-tures. Phys. Rev. Lett. 61, 1650–3 (1988).
[154] Sarma, S. D., de Sousa, R., Hu, X., and Koiller, B. Spin quantum computationin silicon nanostructures. Solid State Commun 133, 737–46 (2005).
[155] Wolf, S. A., Awschalom, D. D., Buhrman, R. A., Daughton, J. M., von Molnar,S., Roukes, M. L., Chtchelkanova, A. Y., and Treger, D. M. Spintronics: aspin-based electronics vision for the future. Science 294, 1488–95 (2001).
167
[156] Kikkawa, J. M. and Awschalom, D. D. Resonant spin amplification in n-typeGaAs. Phys. Rev. Lett. 80, 4313–6 (1998).
[157] Burkard, G., Loss, D., and DiVincenzo, S. P. Coupled quantum dots as quantumgates. Phys. Rev. B 59, 2070–8 (1999).
[158] Ando, T., Fowler, A. B., and Stern, F. Electronic properties of two-dimensionalsystems. Rev. Mod. Phys. 54, 437–672 (1982).
[159] Schulten, K. and Wolynes, P. G. Semi-classical description of electron-spinmotion in radicals including effect of electron hopping. J. Chem. Phys. 68,3292–7 (1978).
[160] Folk, J. A. Spin Transport Measurements in GaAs Quantum Dots. PhD thesis,Stanford University, (2003).
[161] Huibers, A. G. Electron Transport and Dephasing in Semiconductor QuantumDots. PhD thesis, Stanford University, (1999).
168
![Improved Quantum Sensing with a Single Solid-State Spin ...walsworth.physics.harvard.edu/publications/2019_Jaskula_PRApplied.pdfB. Spin-to-charge conversion SCC read-out [12] transfers](https://static.documents.pub/doc/80x56/5eb7b29628a4d83e9c23aa9b/improved-quantum-sensing-with-a-single-solid-state-spin-b-spin-to-charge-conversion.jpg)
